TwitterSentimentAnalysis

Datasets need to be uploaded/called for the .ipynb files to run. ● Datasets:

1. Original Dataset with no emojis a. emotions1 - emotions1.csv - For Baselines, CNN and LSTM b. emotions1 - emotions1.xlsx - For BERT and XLNET
2. Dataset with Emojis as is a. Final_Emojis - emotions1.csv - For Baselines, CNN and LSTM b. Final_Emojis - emotions1.xlsx - For BERT and XLNET
3. Dataset with Emojis converted to words a. Dataset_With_Emoji.csv - For Baselines, CNN and LSTM

● Models:

1. Baseline Models

• Both 1a and 3a should be present locally to get the  output of the Baseline models with and without emojis.
  

a. Baselines - Baseline.ipynb 2. CNN

• Both 1a and 3a should be present locally to get the output of CNN with and without emojis. a. CNN without using word2vec - CNN with word2vec.ipynb b. CNN using word2vec - CNN with word2vec.ipynb

3. LSTM

• The ipynb file is called LSTM.ipynb (this includes all combinations of LSTM embedding layers and is also the file that contains Word2vec+Emoji2vec on CNN) ; Files to be present locally are
  • Emoji2vec.bin
  • Phrase2vec.py
• The 3 datasets mentioned above

4. BERT

• By default emoticons to words mapping is enabled. To disable it, comment cell 11 (apply(token_emoji_conversion) method).
• Mount the drive (helpful in saving checkpoints too) and make sure that dataset is at “/content/gdrive/My Drive/Final_Emojis - emotions1.csv” location.
• Use GPU for code execution
• Specify checkpoint directory under - OUTPUT_DIR in code a. BERT for cased - BERT_converting emoji to word_cased.ipynb; the generated probabilities are saved on the drive in a .csv file named prob_uncased.csv b. BERT for uncased - BERT_converting emoji to word_uncased.ipynb; the generated probabilities are saved on the drive in a .csv file named prob_cased.csv c. BERT(cased+uncased) - read cased and uncased csv files in variables cased_prob and uncased_prob in the code.

5. XLNET

• Mount the drive and make sure that dataset is at the “/content/gdrive/My Drive/Final_Emojis - emotions1.csv” location.
• Specify path from which checkpoint is to be loaded in torch.load
• Use GPU for code execution
• Follow instructions specified in XLNet.pynb notebook. a. For XLNet - XLNet.ipynb