Sentiment Classification

Objective

In this project we trained a pre-built model (BERT) using the transformers library from Hugging Face on a dataset of labeled tweets, labeled for sarcasm (sarcastic/not sarcastic) See link for details on training data. We then used this model to predict the class (sarcastic/not sarcastic) of a set of provided unlabeled tweets for comparison to a competitive baseline score. We were able to beat the baseline with our model. See link for a full report.

Repository Contents

• ./Alternative Methods & Models/: Contains additional models that we built and trained but which were unsuccessful at beating the baseline score.
• ./data/: Contains the test and train data provided for the competition.
• ./Project Documentation/: Contains the final report, demo, and other project deliverables.
• answer.txt: Our final file containing the classification of the test tweets which outperformed the F1 score of the baseline.
• Bert.ipynb: Our notebook file in which we build, train, and test the model.
• TEXT_PREPROCESSING.py: A dependecy of Bert.ipynb, used to preprocess the text for tokenization.