HB

Source code for our paper Hierarchical Blockmodelling for Knowledge Graphs. The implementation is an extension of Qirong Ho's Gibbs Sampler provided to the authors via private correspondence.

Contents

• compile.m defines the necessary paths and the mex compilation and linking function
• run.m is the script to run the Gibbs sampler on the Synthetic Binary Tree dataset described in the paper
• sampler.m is the Matlab wrapper for the C++ Gibbs sampler. This wrapper calls the mex gateway function gateway.cpp
• gateway.ccp is the mex gateway function. It is called by sampler.m and initializes the Gibbs sampler
• hb.hpp is the core of the Gibbs sampler. It defines the tree structure and performs all the necessary sampling computations
• data/synthetic_binary_tree.txt is the for data file the Synthetic Binary Tree dataset. Each line contains the index of the subject, object, and predicate of a triple in the data
• data/fb15k-237.txt is the for data file the FB15k-237 dataset. Each line contains the index of the subject, object, and predicate of a triple in the data
• data/wikidata.txt is the for data file the Wikidata dataset. Each line contains the index of the subject, object, and predicate of a triple in the data
• data/fb15k-237_id2entity.txt is the transaltion file between indices and entities for the FB15k-237 dataset
• data/fb15k-wikidata.txt is the transaltion file between indices and entities for the Wikidata dataset

Installation

1. Install Boost 1.67.0 C++ libraries
2. Install MinGW in MATLAB
3. Update the variables BOOST_PATH and BOOST_LIB_PATH in compile.m to reflect the root and library paths of Boost 1.67.0
4. Run compile.m

Runtime Instructions

Run run.m to perform inference on the model on the Synthetic Binary Tree dataset. If you wish to perform inference using the other two datasets in our paper, follow these steps:

1. Change the size of the input matrix E in compile.m
2. Change the sizes of model_N and model_R in gateway.cpp to correspond to the size of the graph's adjacency tensor
3. Change the input file in the Mex gateway function gateway.cpp
4. Run compile.m to recompile C++ code
5. Update model hyperparameters
6. Run run.m to perform inference