Training different models and checking performance for detecting stance from dataset of Tweets
dataset_train.csv : contains tweets with respect to given target and the Stance labels for that tweet- For, Against or None (Neutral)
dataset_test.csv : contains tweets with respect to given target without any Stance labels
dataset_test_answers.csv : contains the stance labels with respect to tweets as in the test file
DetectingStanceWithBERT.ipynb : Bi-directional Encoder Represenations from Transformers is used for training the data and predicting the test data. The program was run in Google Colab with GPU as runtime
Detecting_Stance_CNN.ipynb.ipynb : CNN has been used for learning the data after pre-processing and forming embedded matrix for each tweet using the pre-trained Word2Vec model. The program was run in Google Colab with CPU as runtime
detecting_stance_svm.ipynb : Here SVM classifier is used to train and predict data before Tf-IDF vectorizer was used to vectorize each word present in vocabulary. The program was run in Jupyter Notebook.
For each of the above approaches, classifiers were trained seperately for each Target.
Performance with BERT was the best among others producing highest F1 Score of 0.63 for Target "Atheism"