CognitiveComputing_MidTerm_SemEval

Sentiment Analysis of SemEval Twitter data and news headlines using Keras for supervised learning http://alt.qcri.org/semeval2017/task5/

Following were the tasks performed (Not in the exact order) for News Headlines:

• Cleaning the data and converting everything to utf-8 format.
• Dividing the sentiments (labels) into 4 categories -1, -0.5, 0.5 and 1 as the values lie between -1 and 1
• We calculate the max length of features and then convert the data into numpy arrays
• We use a convolutional layer for experiment 1(Exp_1), LSTM for the experiment 2 (Exp_2) and convolutional layer along with embedding for the experiment 3 (New Exp 1)
• We use cosine similarity for measuring the performance
• We predict the scores for the training data using the model which we have trained (Accuracy mentioned under model)
• We predict the scores for the test data using the model and display the results

Following were the tasks performed (Not in the exact order) for Twitter Data:

• Cleaning the data and converting everything to utf-8 format.
• Dividing the sentiments (labels) into 4 categories -1, -0.5, 0.5 and 1 as the values lie between -1 and 1 (Hidden you can refer to News Headline for code)
• We calculate the max length of features and then convert the data into numpy arrays
• We use a convolutional layer and LSTM for the experiment
• We use cosine similarity for measuring the performance
• We predict the scores for the training data using the model which we have trained (Accuracy mentioned under model)
• We predict the scores for the test data using the model and display the results