diff --git a/results/260411-kdd-rebuttal-cv-uie-model/PP-UIE-7B_Sculpt-AI_GIMBench-cv-parse_260411-020045.json b/results/260411-kdd-rebuttal-cv-uie-model/PP-UIE-7B_Sculpt-AI_GIMBench-cv-parse_260411-020045.json new file mode 100644 index 0000000..c653276 --- /dev/null +++ b/results/260411-kdd-rebuttal-cv-uie-model/PP-UIE-7B_Sculpt-AI_GIMBench-cv-parse_260411-020045.json @@ -0,0 +1,4323 @@ +{ + "eval_env": { + "exec_command": "/root/autodl-tmp/GIMBench/.venv/bin/python /root/autodl-tmp/GIMBench/src/gimbench/cv/cv_parse.py --use_uie --model_name PP-UIE-7B --api_key **** --base_url https://openrouter.ai/api/v1", + "gimbench_version": "0.4.0", + "gimbench_file": "/root/autodl-tmp/GIMBench/src/gimbench/__init__.py", + "gimkit_version": "0.1.1", + "gimkit_file": "/root/autodl-tmp/GIMBench/.venv/lib/python3.13/site-packages/gimkit/__init__.py", + "git_repo": "/root/autodl-tmp/GIMBench", + "git_branch": "rebuttal/uie-model", + "git_commit_id": "cbbc2739239e411f6f596945be6b77af1ca5b879" + }, + "evaluator_type": "cv", + "args": { + "use_gim_prompt": false, + "output_type": null, + "model_type": null, + "model_name": "PP-UIE-7B", + "api_key": "****", + "base_url": "https://openrouter.ai/api/v1", + "max_model_len": 8192, + "temperature": 0.0, + "top_p": 1.0, + "presence_penalty": 1.0, + "max_tokens": 8192, + "seed": 16, + "first_n": -1, + "num_proc": 1, + "output_dir": "results", + "counter_tokenizer": "Qwen/Qwen3-4B-Instruct-2507", + "record_timing": false, + "ref_model_name": "google/gemma-3-270m", + "ref_model_device": "cpu", + "norm_ppl_alpha": 0.2, + "ppl_window_k": 16, + "golden_truth_only": false, + "no_gimkit": false, + "reason_budget": 0, + "auto_budget": false, + "auto_budget_prompt": "I'll show you a couple of questions. Decide how many reasoning steps are needed to answer each accurately.\n\nConsider a plausible reasoning workflow first (you may use reasoning, reflection, trial and error, and parallel thinking by applying different approaches, plus a quick verification if needed). Then output a step budget (where each step is an atomic reasoning action taking 3–5 sentences) that allows for granular, step-by-step derivation without skipping logic, ensuring a robust and high-confidence conclusion;leave extra headroom for cross-checking and possible revision on multi-hop or tricky questions.\n\n## Question: {question}\n\nDo not be anchored by the examples above. Scale your step budget linearly with the difficulty. For complex problems, you are encouraged to assign a high budget (20, or more) to ensure there is enough room for step-by-step derivation and verification.\n\n", + "reason_step_desc": "A distinct, verified reasoning step building on the previous one. Write 2–3 substantial sentences (60–80 words each) to ensure depth.", + "use_outlines": false, + "use_uie": true, + "judge_model_name": "google/gemini-3-flash-preview", + "num_samples": 1, + "pass_k": [ + 1 + ], + "exec_timeout": 10, + "scierc_split": "test", + "scierc_max_relations": 30, + "dataset": { + "path": "Sculpt-AI/GIMBench-cv-parse", + "name": null, + "split": "test" + } + }, + "start_time": "2026-04-11T02:00:45.143288", + "end_time": "2026-04-11T02:18:04.299160", + "elapsed_minutes": 17.319264533333335, + "evaled_items": [ + { + "filename": "Nan_Zhang.pdf", + "extraction_details": { + "name": { + "prediction": "Nan Zhang", + "expected": "Nan Zhang", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "Italy", + "expected": "Germany", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "3 January 1984", + "expected": "1984-01-03", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "phone_number": { + "prediction": "+ 4 9$ (0) 621 181 2098", + "expected": "+49 (0) 621 181 2098", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "email": { + "prediction": "b n.zhang@uni-mannheim.de", + "expected": "n.zhang@uni-mannheim.de", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Stanford University", + "expected": "Stanford University", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Max-Planck-Institute for Research on Collective Goods", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Economics", + "expected": "Political Science", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "2014", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2012", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "b", + "expected": "http://nanzhangresearch.github.io", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "nanzhangresearch", + "expected": "https://github.com/nanzhangresearch", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + } + }, + "num_fields": 10, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "Aurelie_de_Gendre.pdf", + "extraction_details": { + "name": { + "prediction": "Andrew Mallos", + "expected": "Aurelie de Gendre", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "France", + "expected": "France", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "2022-2027", + "expected": "1990-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+33", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "Nicolás Salamanca", + "expected": "a.degendre@unimelb.edu.au", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Humboldt University Berlin", + "expected": "Maastricht University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Department of Economics", + "expected": "Research Centre for Education and the Labor Market", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Statistics", + "expected": "Economics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "2011-2014", + "expected": "2015-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2021", + "expected": "2021-01-01", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "homepage_url": { + "prediction": "France", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "Andrew Mallos", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 10, + "num_correct": 3, + "error_msg": "" + }, + { + "filename": "Aditya_Chaudhry.pdf", + "extraction_details": { + "name": { + "prediction": "Aditya Chaudhry", + "expected": "Aditya Chaudhry", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "United States of America", + "expected": "United States of America", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "29", + "expected": "1996-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "无相应实体", + "expected": "703 628 9071", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "AdityaChaudhry", + "expected": "chaudhry.127@osu.ed", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Chicago", + "expected": "University of Chicago", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Fisher College of Business", + "expected": "Booth School of Business", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Finance", + "expected": "Finance", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "2018", + "expected": "2018-01-01", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "end_date": { + "prediction": "2028", + "expected": "2023-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "无相应实体", + "expected": "https://chaudhryaditya.github.io/", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "Aditya Chaudhry", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "Alexander_J_Fertig.pdf", + "extraction_details": { + "name": { + "prediction": "Alexander J. Fertig", + "expected": "Alexander J. Fertig", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "U.S.", + "expected": "United States", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "birthday": { + "prediction": "1990", + "expected": "1988-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "(716) 587-1812", + "expected": "(716) 587-1812", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "email": { + "prediction": "afertig@umich.edu", + "expected": "afertig@umich.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Michigan", + "expected": "University of Michigan", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Department of Economics", + "expected": "Department of Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "major": { + "prediction": "Mathematics-Economics", + "expected": "Economics & Public Policy", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "2010", + "expected": "2019-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2010", + "expected": "2026-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "alexanderfertig.com", + "expected": "http://alexanderfertig.com", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "alexanderfertig.com", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 8, + "error_msg": "" + }, + { + "filename": "Maksim_Zhdanov.pdf", + "extraction_details": { + "name": { + "prediction": "MAKSIM ZHDANOV", + "expected": "Maksim Zhdanov", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "Quebec", + "expected": "Netherlands", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "22", + "expected": "1997-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+ +$", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "MAKSIM ZHDANOV", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "PhD", + "expected": "Master", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "university": { + "prediction": "University of Amsterdam", + "expected": "University of Amsterdam", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "AI", + "expected": "AMLab", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Physics", + "expected": "Machine Learning", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "Apr 22, 222", + "expected": "2023-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "Apr 22", + "expected": "2027-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "twitter", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "MAKSIM ZHDANOV", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 9, + "num_correct": 2, + "error_msg": "" + }, + { + "filename": "Theo_Serlin.pdf", + "extraction_details": { + "name": { + "prediction": "Theo Serlin", + "expected": "Theo Serlin", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "UK", + "expected": "United Kingdom", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "birthday": { + "prediction": "11, 22025", + "expected": "1997-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "theo-serlin.github.io", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "theo.serlin@kcl.ac.uk", + "expected": "theo.serlin@kcl.ac.uk", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Harvard University", + "expected": "Stanford University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Political Science", + "expected": "Department of Political Science", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "major": { + "prediction": "Economics", + "expected": "Political Science", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "22", + "expected": "2019-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2222", + "expected": "2024-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "theo-serlin.github.io", + "expected": "https://theo-serlin.github.io", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "theo-serlin.github.io", + "expected": "https://github.com/theo-serlin", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "August_Bruno.pdf", + "extraction_details": { + "name": { + "prediction": "August Bruno", + "expected": "August Bruno", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "United States", + "expected": "United States", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "May 2, 2217", + "expected": "1995-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "(716) 2", + "expected": "+1 (716) 238-5546", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "abruno@skidmore.edu", + "expected": "abruno@skidmore.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of North Carolina at Chapel Hill", + "expected": "University of North Carolina at Chapel Hill", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Department of Economics", + "expected": "Department of Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "major": { + "prediction": "Economics", + "expected": "Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "May 2, 2217", + "expected": "2025-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "May 2, 2217", + "expected": "2025-05-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "augustbruno.github.io", + "expected": "https://augustbruno.github.io", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "augustbruno", + "expected": "https://github.com/augustbruno", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + } + }, + "num_fields": 13, + "num_correct": 9, + "error_msg": "" + }, + { + "filename": "Unknown.pdf", + "extraction_details": { + "name": { + "prediction": "University of U", + "expected": "Amarnath Murugan", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "U", + "expected": "USA", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "EDUCATION", + "expected": "1997-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "University of U", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "EDUCATION", + "expected": "amarnathmurugan0@gmail.com", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "MS", + "expected": "Master", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of U", + "expected": "University of Utah", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "MS", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Computing, Graphics & Visualization", + "expected": "Graphics & Visualization", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "EDUCATION", + "expected": "2022-08-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "MS", + "expected": "2024-05-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "University of U", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "University of U", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 9, + "num_correct": 1, + "error_msg": "" + }, + { + "filename": "Guillermo_A_Perez.pdf", + "extraction_details": { + "name": { + "prediction": "Guillermo A. Perez´", + "expected": "Guillermo A. Perez", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "Belgium", + "expected": "Belgium", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "2020", + "expected": "1987-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+3232653904", + "expected": "+3232653904", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "email": { + "prediction": "guillermo.perez@uantwerpen.be", + "expected": "guillermo.perez@uantwerpen.be", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Université libre de Bruxelles", + "expected": "Université libre de Bruxelles", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Computer Science", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Computer Science", + "expected": "Computer Science", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "2018-2021", + "expected": "2012-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2020-2023", + "expected": "2016-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "www.uantwerpen.be/en/staff/guillermoalberto-perez/", + "expected": "https://www.uantwerpen.be/en/staff/guillermoalberto-perez/", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "www.uantwerpen.be/en/staff/guillermoalberto-perez/", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 8, + "error_msg": "" + }, + { + "filename": "Nicholas_Vreugdenhil.pdf", + "extraction_details": { + "name": { + "prediction": "Nicholas Vreugdenhil", + "expected": "Nicholas Vreugdenhil", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "Australia", + "expected": "Australia", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "22", + "expected": "1990-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "Nicholas Vreugdenhil", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "nvreugde@asu.edu", + "expected": "nvreugde@asu.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Arizona State University", + "expected": "Northwestern University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "W.P. Carey School of Business", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "E", + "expected": "Economics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "July 22, 22020", + "expected": "2020-07-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2029", + "expected": "2026-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "https://nvreug.github.io", + "expected": "https://nvreug.github.io", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "github_url": { + "prediction": "https://nvreug.github.io", + "expected": "https://github.com/nvreug", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 5, + "error_msg": "" + }, + { + "filename": "Florian_P_Ederer.pdf", + "extraction_details": { + "name": { + "prediction": "", + "expected": "Florian P. Ederer", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "", + "expected": "USA", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "", + "expected": "1980-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "", + "expected": "florian.ederer@gmail.com", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "university": { + "prediction": "", + "expected": "Massachusetts Institute of Technology", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "", + "expected": "Department of Economics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "", + "expected": "Economics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "", + "expected": "2004-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "", + "expected": "2009-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "", + "expected": "https://florianederer.github.io/", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "", + "expected": "https://github.com/florianederer", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 0, + "error_msg": "" + }, + { + "filename": "Ivan_Kartáč.pdf", + "extraction_details": { + "name": { + "prediction": "Ivan Kart´aˇc", + "expected": "Ivan Kartáč", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "Czechia", + "expected": "Czechia", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "10/22", + "expected": "1996-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+42", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "kartac@ufal.mff.cuni.cz", + "expected": "kartac@ufal.mff.cuni.cz", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Bachelor's degree", + "expected": "Master", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "university": { + "prediction": "Charles University, Faculty of Mathematics and Physics", + "expected": "Charles University", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "department": { + "prediction": "Faculty of Mathematics and Physics", + "expected": "Faculty of Mathematics and Physics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "major": { + "prediction": "General Linguistics and Phonetics", + "expected": "Language Technologies and Computational Linguistics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "07/2225", + "expected": "2021-10-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "02/2025", + "expected": "2025-02-01", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "homepage_url": { + "prediction": "kartac@ufal.mff.cuni.cz", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "ivan-kartac", + "expected": "https://github.com/ivan-kartac", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + } + }, + "num_fields": 11, + "num_correct": 7, + "error_msg": "" + }, + { + "filename": "Diego_Martínez.pdf", + "extraction_details": { + "name": { + "prediction": "Diego Mart´ınez", + "expected": "Diego Martínez", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "Belgium", + "expected": "Spain", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "19", + "expected": "1992-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "diego.martinez@kuleuven.be", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "diego.martinez@kuleuven.be", + "expected": "diego.martinez@kuleuven.be", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "KU Leuven", + "expected": "Universidad Carlos III de Madrid", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Mathematics", + "expected": "Department of Mathematics", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "major": { + "prediction": "Mathematics", + "expected": "Mathematics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "September 2024", + "expected": "2025-10-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "September 2024", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "https://diego-mmg.github.io/", + "expected": "https://diego-mmg.github.io/", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "github_url": { + "prediction": "diego-mmg.github.io", + "expected": "https://github.com/diego-mmg", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "Yuki_Takahashi.pdf", + "extraction_details": { + "name": { + "prediction": "", + "expected": "Yuki Takahashi", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "", + "expected": "The Netherlands", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "", + "expected": "1987-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "", + "expected": "+31 6-4145-8415", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "", + "expected": "y.takahashi@tilburguniversity.edu", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "university": { + "prediction": "", + "expected": "University of Bologna", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "", + "expected": "Dept. of Economics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "", + "expected": "Economics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "", + "expected": "2017-11-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "", + "expected": "2022-07-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "", + "expected": "https://yukitakahashi1.github.io", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "", + "expected": "https://github.com/yukitakahashi1", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 13, + "num_correct": 0, + "error_msg": "" + }, + { + "filename": "Richard_Cornelius_Suwandi.pdf", + "extraction_details": { + "name": { + "prediction": "Richard Cornelius Suwandi", + "expected": "Richard Cornelius Suwandi", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "Shenzhen", + "expected": "China", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "1997", + "expected": "2001-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "richardsuwandi@link.cuhk.edu.cn", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "richardsuwandi@link.cuhk.edu.cn", + "expected": "richardsuwandi@link.cuhk.edu.cn", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "PhD", + "expected": "PhD", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "university": { + "prediction": "The Chinese University of Hong Kong, Shenzhen", + "expected": "The Chinese University of Hong Kong, Shenzhen", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Computer and Information Engineering", + "expected": "School of Science and Engineering", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Statistics (Data Science Stream)", + "expected": "Computer and Information Engineering", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "Sep 22, 2969", + "expected": "2023-09-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "May 22, 29", + "expected": "2027-05-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "richardcsuwandi.github.io", + "expected": "https://richardcsuwandi.github.io", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "richardcsuwandi.github.io", + "expected": "https://github.com/richardcsuwandi", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 5, + "error_msg": "" + }, + { + "filename": "Wojciech_Anyszka.pdf", + "extraction_details": { + "name": { + "prediction": "Wojciech Anyszka", + "expected": "Wojciech Anyszka", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "Saudi Arabia", + "expected": "USA", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "2", + "expected": "2001-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "531 333 131", + "expected": "531 333 131", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "email": { + "prediction": "wa8117@princeton.edu", + "expected": "wa8157@princeton.edu", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "PhD", + "expected": "PhD", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "university": { + "prediction": "University of Groningen", + "expected": "Princeton University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Operations Research and Financial Engineering (ORFE)", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Mathematical Sciences", + "expected": "Operations Research and Financial Engineering", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "Sept 22, 22", + "expected": "2025-09-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "Nov 22, 22/23", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "wanyszka.github.io", + "expected": "https://wanyszka.github.io", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "wanyszka.github.io", + "expected": "https://github.com/wanyszka", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 4, + "error_msg": "" + }, + { + "filename": "A_Combs.pdf", + "extraction_details": { + "name": { + "prediction": "Combs", + "expected": "A. Combs", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "USA", + "expected": "United States", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "birthday": { + "prediction": "2009", + "expected": "1987-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "无相应实体", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "email": { + "prediction": "Combs", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Kentucky", + "expected": "University of Kentucky", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Public Administration", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Public Administration", + "expected": "Public Policy & Administration", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "start_date": { + "prediction": "2019-2020", + "expected": "2014-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2020-Present", + "expected": "2018-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "University of Kentucky", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "Combs", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 8, + "num_correct": 5, + "error_msg": "" + }, + { + "filename": "Elena_Pagnin.pdf", + "extraction_details": { + "name": { + "prediction": "ELENA PAGNIN", + "expected": "Elena Pagnin", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "Swedish", + "expected": "Italy", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "2011-2013", + "expected": "1989-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+46704704704", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "elenap@chalmers.se", + "expected": "elenap@chalmers.se", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Chalmers", + "expected": "Chalmers University of Technology", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "department": { + "prediction": "Department of Computer Science and Engineering", + "expected": "Department of Computer Science and Engineering", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "major": { + "prediction": "Pure Mathematics", + "expected": "Computer Science", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "2022 Oct-now", + "expected": "2014-05-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2011-2013", + "expected": "2019-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "epagnin.github.io", + "expected": "https://epagnin.github.io", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "epagnin.github.io", + "expected": "https://github.com/epagnin", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "Gabriel_Okasa.pdf", + "extraction_details": { + "name": { + "prediction": "Gabriel Okasa", + "expected": "Gabriel Okasa", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "Switzerland", + "expected": "Switzerland", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "1990", + "expected": "1992-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+41 78 700 07 63", + "expected": "+41 78 700 07 63", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "email": { + "prediction": "gabriel.okasa@epfl.ch", + "expected": "gabriel.okasa@epfl.ch", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Bachelor Degree", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "university": { + "prediction": "University of St.Gallen", + "expected": "University of St.Gallen", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "E", + "expected": "School of Economics and Political Science", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Management", + "expected": "Economics and Finance", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "February 22, 218", + "expected": "2017-02-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "February 22, 22", + "expected": "2022-02-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "okasag.github.io", + "expected": "https://okasag.github.io", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "okasag.github.io", + "expected": "https://github.com/okasag", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + } + }, + "num_fields": 13, + "num_correct": 7, + "error_msg": "" + }, + { + "filename": "Martin_Souchier.pdf", + "extraction_details": { + "name": { + "prediction": "Martinsouchier", + "expected": "Martin Souchier", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "France", + "expected": "USA", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "29", + "expected": "1993-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "souchier@wharton.upenn.edu", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "souchier@wharton.upenn.edu", + "expected": "souchier@wharton.upenn.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Pennsylvania", + "expected": "Stanford University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Wharton School, University of Pennsylvania", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Economics", + "expected": "Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "2011 - 2014", + "expected": "2017-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "29", + "expected": "2023-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "souchier@wharton.upenn.edu", + "expected": "https://www.martinsouchier.com/", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "souchier@wharton.upenn.edu", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 10, + "num_correct": 4, + "error_msg": "" + }, + { + "filename": "Alexey_Makarin.pdf", + "extraction_details": { + "name": { + "prediction": "", + "expected": "Alexey Makarin", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "", + "expected": "Russia", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "", + "expected": "1990-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "", + "expected": "(224) 304-3896", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "", + "expected": "makarin@mit.edu", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "university": { + "prediction": "", + "expected": "Northwestern University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "", + "expected": "Economics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "", + "expected": "2014-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "", + "expected": "2019-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "", + "expected": "https://alexymakarin.github.io/", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "", + "expected": "https://github.com/alexymakarin", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 0, + "error_msg": "" + }, + { + "filename": "Minguk_Kang.pdf", + "extraction_details": { + "name": { + "prediction": "Minginguk Kang", + "expected": "Minguk Kang", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "Republic of Korea", + "expected": "Republic of Korea", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "16th", + "expected": "1994-03-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "77", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "mgkang@postech.ac.kr", + "expected": "mgkang@postech.ac.kr", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Pusan National University", + "expected": "POSTECH", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Computer Vision Laboratory", + "expected": "Graduate School of AI", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Mechanical Engineering, Minor: Statistics", + "expected": "AI", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "Mar 22013", + "expected": "2020-02-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "Mar 22, 22", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "mgkang@postech.ac.kr", + "expected": "https://mingukkang.github.io", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "GitHub", + "expected": "https://github.com/POSTECH-CVL/PyTorch-StudioGAN", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 3, + "error_msg": "" + }, + { + "filename": "Not_Provided.pdf", + "extraction_details": { + "name": { + "prediction": "Naomi Hirano", + "expected": "Yu-Hsuan Teng", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "Taiwan", + "expected": "Taiwan", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "1995", + "expected": "1995-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "30", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "无相应实体", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "highest_level_degree": { + "prediction": "M. S.", + "expected": "Master", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "National Taiwan University", + "expected": "National Taiwan University", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Physics", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Electrical Engineering", + "expected": "Physics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "Sep. 2213", + "expected": "2017-09-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "Jun. 2017", + "expected": "2019-06-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "National Taiwan University", + "expected": "https://elthateng.github.io", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "无相应实体", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + } + }, + "num_fields": 9, + "num_correct": 5, + "error_msg": "" + }, + { + "filename": "Jason_Alan_Snyder.pdf", + "extraction_details": { + "name": { + "prediction": "Jason Alan Snyder", + "expected": "Jason Alan Snyder", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "USA", + "expected": "USA", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "26", + "expected": "1978-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "(310)-210-1471", + "expected": "310-210-1471", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "email": { + "prediction": "jsnyder@eccles.utah.edu", + "expected": "jsnyder@eccles.utah.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Utah", + "expected": "University of California, Berkeley", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Management Science", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Applied Mathematics", + "expected": "Business Administration", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "26", + "expected": "2000-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "26", + "expected": "2006-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "无相应实体", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "无相应实体", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + } + }, + "num_fields": 10, + "num_correct": 7, + "error_msg": "" + }, + { + "filename": "Zsigmond_Pálvölgyi.pdf", + "extraction_details": { + "name": { + "prediction": "Zsigmond Pálvölgyi", + "expected": "Zsigmond Pálvölgyi", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "Hungary", + "expected": "Hungary", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "December 2", + "expected": "1991-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "zsigmond.palvolgyi@uni-corvinus.hu", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "zsigmond.palvolgyi@uni-corvinus.hu", + "expected": "zsigmond.palvolgyi@uni-corvinus.hu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Corvinus University of Budapest", + "expected": "University of Michigan", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Economics", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Economics", + "expected": "Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "2023", + "expected": "2018-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2013-14", + "expected": "2025-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "https://zsigmond-palvolgyi.github.io/", + "expected": "https://zsigmond-palvolgyi.github.io/", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "github_url": { + "prediction": "zsigmond-palvolgyi.github.io/", + "expected": "https://github.com/zsigmond-palvolgyi", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "Mayara_Felix.pdf", + "extraction_details": { + "name": { + "prediction": "MAYARA FELIX", + "expected": "Mayara Felix", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "Brazil", + "expected": "Brazil", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "1990", + "expected": "1989-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+1 413 437 0105", + "expected": "+1 413 437 0105", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "email": { + "prediction": "mayara.felix@yale.edu", + "expected": "mayara.felix@yale.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Yale University", + "expected": "Massachusetts Institute of Technology", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Economics", + "expected": "Department of Economics", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "major": { + "prediction": "Economics", + "expected": "Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "2021-2022", + "expected": "2015-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "May 2, 2211", + "expected": "2021-06-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "www.mayarafelix.com", + "expected": "http://www.mayarafelix.com", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "www.mayarafelix.com", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 8, + "error_msg": "" + }, + { + "filename": "Filippo_Pallotti.pdf", + "extraction_details": { + "name": { + "prediction": "Filippo Pallotti", + "expected": "Filippo Pallotti", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "Italy", + "expected": "Italy", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "2010-22", + "expected": "1991-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+393384892976", + "expected": "+393384892976", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "email": { + "prediction": "filippo.pallotti@ucl.ac.uk", + "expected": "filippo.pallotti@ucl.ac.uk", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "PhD", + "expected": "PhD", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "university": { + "prediction": "University College London", + "expected": "University College London", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Department of Economics", + "expected": "Department of Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "major": { + "prediction": "Economics", + "expected": "Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "Nov. 2023", + "expected": "2019-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2023", + "expected": "2025-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "www.filippopallotti.com", + "expected": "http://www.filippopallotti.com", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "www.filippopallotti.com", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 9, + "error_msg": "" + }, + { + "filename": "Hazel_Doughty.pdf", + "extraction_details": { + "name": { + "prediction": "Hazel Doughty", + "expected": "Hazel Doughty", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "Netherlands", + "expected": "Netherlands", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "Dec 2020", + "expected": "1994-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "h.r.doughty@liacs.leidenuniv.nl", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "h.r.doughty@liacs.leidenuniv.nl", + "expected": "h.r.doughty@liacs.leidenuniv.nl", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "PhD", + "expected": "PhD", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "university": { + "prediction": "Leiden University", + "expected": "University of Bristol", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "LIACS, Leiden University", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Computer Science", + "expected": "Computer Science", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "Jan 2019", + "expected": "2016-09-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "June 2016", + "expected": "2020-09-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "hazeldoughty.github.io", + "expected": "https://hazeldoughty.github.io", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "hazeldoughty.github.io", + "expected": "https://github.com/hazeldoughty", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "Masoud_Jasbi.pdf", + "extraction_details": { + "name": { + "prediction": "Masoud Jasbi", + "expected": "Masoud Jasbi", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "USA", + "expected": "United States", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "birthday": { + "prediction": "2024", + "expected": "1990-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+1 530-754-1111", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "jasbi@ucdavis.edu", + "expected": "jasbi@ucdavis.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of California", + "expected": "Stanford University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Department of Linguistics", + "expected": "Department of Linguistics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "major": { + "prediction": "Linguistics", + "expected": "Linguistics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "2012-18", + "expected": "2012-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2020", + "expected": "2018-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "https://jasbi.github.io", + "expected": "https://jasbi.github.io", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "github_url": { + "prediction": "https://jasbi.github.io", + "expected": "https://github.com/jasbi", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 7, + "error_msg": "" + }, + { + "filename": "Nikola_Konstantinov.pdf", + "extraction_details": { + "name": { + "prediction": "DR. NIKOLA KONSTANTINOV", + "expected": "Nikola Konstantinov", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "Bulgaria", + "expected": "Bulgaria", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "2005", + "expected": "1994-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+35921315111", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "nikola.konstantinov@insait.ai", + "expected": "nikola.konstantinov@insait.ai", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "PhD", + "expected": "PhD", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "university": { + "prediction": "ETH Zurich", + "expected": "IST Austria", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "IST Austria", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Mathematics and Statistics", + "expected": "Machine Learning", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "06.2023", + "expected": "2017-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "04.2023", + "expected": "2022-03-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "https://insait.ai/dr-nikola-konstantinov", + "expected": "https://insait.ai/dr-nikola-konstantinov", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "github_url": { + "prediction": "strategic-fl-sim", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 10, + "num_correct": 5, + "error_msg": "" + }, + { + "filename": "Devin_Judge-Lord.pdf", + "extraction_details": { + "name": { + "prediction": "Devin Judge-Lord", + "expected": "Devin Judge-Lord", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "U.S.", + "expected": "United States", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "birthday": { + "prediction": "1990", + "expected": "1987-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+1", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "judgelor@umich.edu", + "expected": "judgelor@umich.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Michigan", + "expected": "University of Wisconsin–Madison", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Department of Government", + "expected": "Department of Political Science", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "major": { + "prediction": "Political Science", + "expected": "Political Science", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "2021-2023", + "expected": "2023-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2021", + "expected": "2025-04-28", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "judgelord.github.io", + "expected": "https://judgelord.github.io", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "judgelord.github.io", + "expected": "https://github.com/judgelord", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 7, + "error_msg": "" + }, + { + "filename": "Amanda_R_Kreider.pdf", + "extraction_details": { + "name": { + "prediction": "Amanda R. Kreider", + "expected": "Amanda R. Kreider", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "United States", + "expected": "United States", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "1990", + "expected": "1987-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "130 De Soto Street", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "amandakreider@pitt.edu", + "expected": "amandakreider@pitt.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Pennsylvania", + "expected": "Harvard University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Health Policy & Management", + "expected": "Department of Health Policy & Management", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "major": { + "prediction": "Economics", + "expected": "Health Policy, Economics Track", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "2015–2016", + "expected": "2015-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2024", + "expected": "2021-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "130 De Soto Street", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "无相应实体", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + } + }, + "num_fields": 10, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "K_Zhu.pdf", + "extraction_details": { + "name": { + "prediction": "Zhu", + "expected": "K. Zhu", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "USA", + "expected": "China", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "2015", + "expected": "1997-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "UIUC", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "Zhu", + "expected": "kunlunz2@illinois.edu", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "Master of Science", + "expected": "Master", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Illinois Urbana Chanpaign", + "expected": "University of Illinois Urbana Chanpaign", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "CS", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Computer Science", + "expected": "Computer Science", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "Aug 2024", + "expected": "2024-08-01", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "end_date": { + "prediction": "May 2026", + "expected": "2026-05-01", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "homepage_url": { + "prediction": "arXiv", + "expected": "https://kzhu.github.io", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "ToolBench", + "expected": "https://github.com/Kunlun-Zhu", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 5, + "error_msg": "" + }, + { + "filename": "Matthieu_Blanke.pdf", + "extraction_details": { + "name": { + "prediction": "Matthieu Blanke", + "expected": "Matthieu Blanke", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "France", + "expected": "France", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "10/22", + "expected": "1999-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+33", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "Matthieu Blanke", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "NYU Courant Institute of Mathematical Sciences", + "expected": "Inria Paris and DI ENS", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Mathématiques, Vision, Apprentissage", + "expected": "DI ENS", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Applied mathematics, computer science, theoretical physics", + "expected": "Deep learning for physical systems", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "22", + "expected": "2021-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "22", + "expected": "2024-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "GitHub", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "Matthieu Blanke", + "expected": "https://github.com/MatthieuBlanke", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 10, + "num_correct": 3, + "error_msg": "" + }, + { + "filename": "Paul_G_Freed.pdf", + "extraction_details": { + "name": { + "prediction": "PAUL G. FREED", + "expected": "Paul G. Freed", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "Cyprus", + "expected": "United States", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "January 8th, 2", + "expected": "1998-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "freedp@darden.virginia.edu", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "freedp@darden.virginia.edu", + "expected": "freedp@darden.virginia.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Virginia, Darden School of Business", + "expected": "University of South Carolina", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Business Administration, Finance Area", + "expected": "Darla Moore School of Business", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Business Administration, Finance Area", + "expected": "Finance", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "start_date": { + "prediction": "2224", + "expected": "2020-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2224", + "expected": "2025-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "freedp@darden.virginia.edu", + "expected": "https://freedp.github.io/", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "freedp@darden.virginia.edu", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 4, + "error_msg": "" + }, + { + "filename": "Eric_J_Leonardis.pdf", + "extraction_details": { + "name": { + "prediction": "Eric J. Leonardis", + "expected": "Eric J. Leonardis", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "USA", + "expected": "United States", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "birthday": { + "prediction": "110", + "expected": "1992-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "516-510-2096", + "expected": "516-510-2096", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "email": { + "prediction": "Eric J. Leonardis", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of California, San Diego", + "expected": "University of California, San Diego", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Cognitive Science", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Cognitive Science – Systems Neuroscience, Robotics", + "expected": "Systems Neuroscience, Robotics", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "start_date": { + "prediction": "Fall 2", + "expected": "2014-09-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "Spring 2", + "expected": "2022-06-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "Eric J. Leonardis, MS, PhD", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "Eric J. Leonardis", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 9, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "Justin_Junyang_Lin.pdf", + "extraction_details": { + "name": { + "prediction": "Justin (Junyang) Lin", + "expected": "Justin (Junyang) Lin", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "Japan", + "expected": "China", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "19", + "expected": "1994-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "11", + "expected": "+86 156-5296-8896", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "Junyang Lin", + "expected": "linjunyang@pku.edu.cn", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "Bachelor of English Literature", + "expected": "Master", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "university": { + "prediction": "University of International Relations", + "expected": "Peking University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Institute of Computational Linguistics", + "expected": "Institute of Linguistics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Langu", + "expected": "Linguistics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "July 2", + "expected": "2016-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "July 2219", + "expected": "2019-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "无相应实体", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "无相应实体", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + } + }, + "num_fields": 11, + "num_correct": 3, + "error_msg": "" + }, + { + "filename": "Xiaoyang_Ye.pdf", + "extraction_details": { + "name": { + "prediction": "Xiaoyang Ye", + "expected": "Xiaoyang Ye", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "China", + "expected": "USA", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "December 2025", + "expected": "1988-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+1", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "xiaoyang.ye26@gmail.com", + "expected": "xiaoyang.ye26@gmail.com", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "Brown University", + "expected": "University of Michigan", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Economics", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Economics", + "expected": "Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "December 2025", + "expected": "2014-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "December 2025", + "expected": "2019-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "https://xiaoyangye.github.io/", + "expected": "https://xiaoyangye.github.io/", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "github_url": { + "prediction": "https://xiaoyangye.github.io/", + "expected": "https://github.com/xiaoyangye", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 11, + "num_correct": 5, + "error_msg": "" + }, + { + "filename": "Pedro_Rodriguez-Martinez.pdf", + "extraction_details": { + "name": { + "prediction": "Pedro Rodriguez-Martinez", + "expected": "Pedro Rodriguez-Martinez", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "USA", + "expected": "USA", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "2011", + "expected": "1993-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "+1 (628) 252-8644", + "expected": "+1 (628) 252-8644", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "email": { + "prediction": "perodmar@gwu.edu", + "expected": "perodmar@gwu.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "PhD", + "expected": "PhD", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "university": { + "prediction": "George Washington University", + "expected": "George Washington University", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Economics", + "expected": "Department of Economics", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "major": { + "prediction": "Economics", + "expected": "Economics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "08/22", + "expected": "2020-08-01", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "end_date": { + "prediction": "05/22", + "expected": "2026-05-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "perodmar.github.io", + "expected": "https://perodmar.github.io", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "github_url": { + "prediction": "perodmar.github.io", + "expected": "https://github.com/perodmar", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + } + }, + "num_fields": 13, + "num_correct": 11, + "error_msg": "" + }, + { + "filename": "Yifan_Jiang.pdf", + "extraction_details": { + "name": { + "prediction": "Yifan Jiang", + "expected": "Yifan Jiang", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "country": { + "prediction": "United Kingdom", + "expected": "United Kingdom", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "birthday": { + "prediction": "1997", + "expected": "1998-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "yifan.jiang@imperial.ac.uk", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "yifan.jiang@imperial.ac.uk", + "expected": "yifan.jiang@imperial.ac.uk", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "DPhil", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of Oxford", + "expected": "University of Oxford", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Mathematics", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Mathematics", + "expected": "Mathematics", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "start_date": { + "prediction": "Sep 2025", + "expected": "2021-09-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2025-10-18", + "expected": "2025-09-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "无相应实体", + "expected": "https://yifanjiang233.github.io", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "yifanjiang233", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 10, + "num_correct": 6, + "error_msg": "" + }, + { + "filename": "ZHOU_HUACHI.pdf", + "extraction_details": { + "name": { + "prediction": "HUACHI, ZHOU, Huachi, Zhou, Huachi, Zhou, Huachi, Zhou", + "expected": "Huachi Zhou", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "country": { + "prediction": "CHINA", + "expected": "Hong Kong", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "09/22", + "expected": "1998-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "60629081", + "expected": "60629081", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "email": { + "prediction": "huachi.zhou@connect.polyu.hk", + "expected": "huachi.zhou@connect.polyu.hk", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Bachelor of Software Engineering", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "university": { + "prediction": "East China Normal University", + "expected": "The Hong Kong Polytechnic University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "Engineering", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Master of Data Science", + "expected": "Computing", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "09/22", + "expected": "2022-09-01", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "end_date": { + "prediction": "07/2019", + "expected": "2026-09-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "connect.polyu.hk", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "huachi.zhou", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 10, + "num_correct": 4, + "error_msg": "" + }, + { + "filename": "Zarek_Brot-Goldberg.pdf", + "extraction_details": { + "name": { + "prediction": "zarek", + "expected": "Zarek Brot-Goldberg", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "U.S.", + "expected": "USA", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "birthday": { + "prediction": "1307 E 60th St., #3011", + "expected": "1989-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "607-606-37", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "zarek@uchicago.edu", + "expected": "zarek@uchicago.edu", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "highest_level_degree": { + "prediction": "Ph.D.", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": true, + "correct": true + }, + "university": { + "prediction": "University of California at Berkeley", + "expected": "University of California at Berkeley", + "verbatim_correct": true, + "judge_model_correct": false, + "correct": true + }, + "department": { + "prediction": "Economics", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "Mathematics", + "expected": "Economics", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "2020", + "expected": "2013-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "2020", + "expected": "2019-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "zarekcb", + "expected": "http://sites.google.com/site/zarekcb", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "zarekcb", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 10, + "num_correct": 4, + "error_msg": "" + }, + { + "filename": "Zubin_Jelveh.pdf", + "extraction_details": { + "name": { + "prediction": "", + "expected": "Zubin Jelveh", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "country": { + "prediction": "", + "expected": "USA", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "birthday": { + "prediction": "", + "expected": "1977-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "phone_number": { + "prediction": "", + "expected": null, + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "email": { + "prediction": "", + "expected": "zjelveh@umd.edu", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "highest_level_degree": { + "prediction": "", + "expected": "PhD", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "university": { + "prediction": "", + "expected": "New York University", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "department": { + "prediction": "", + "expected": "Tandon School of Engineering", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "major": { + "prediction": "", + "expected": "Computer Science", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "start_date": { + "prediction": "", + "expected": "2011-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "end_date": { + "prediction": "", + "expected": "2017-01-01", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "homepage_url": { + "prediction": "", + "expected": "http://zjelveh.github.io", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + }, + "github_url": { + "prediction": "", + "expected": "https://github.com/zjelveh", + "verbatim_correct": false, + "judge_model_correct": false, + "correct": false + } + }, + "num_fields": 12, + "num_correct": 0, + "error_msg": "" + } + ], + "total_fields": 472, + "total_correct": 218, + "errors": 0, + "calibrated_accuracy": 0.461864406779661 +} diff --git a/results/260411-kdd-rebuttal-cv-uie-model/eval.sh b/results/260411-kdd-rebuttal-cv-uie-model/eval.sh new file mode 100644 index 0000000..bca74f7 --- /dev/null +++ b/results/260411-kdd-rebuttal-cv-uie-model/eval.sh @@ -0,0 +1,6 @@ +export API_KEY=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 +export API_BASE=https://openrouter.ai/api/v1 + +python -m gimbench.cv.cv_parse --use_uie --model_name "PP-UIE-7B" --api_key $API_KEY --base_url $API_BASE + +shutdown -h +3