Requirements

• pytorch == 1.8.1
• transformers == 4.2.2
• pytorch-pretrained-bert==0.6.2
• sklearn == 0.21
• scikit-learn==0.23.2
• scipy==1.5.4
• imblearn==0.0
• numpy == 1.19.5
• pandas == 1.1.5

Data CSV Format

| DialogueId | UtteranceId | Speaker | Utterance | Persona? | Persona Type |

Training and Evaluation

• Download and place 200d GloVe word vectors in 'data/'folder
• Prepare data in the given format and place it in 'data/' folder.
• Execution
  • Persona Discovery - python dataloader.py. It prepares and dumps the dataframes in a format which can be used by models and dataloaders to train and evaluate.
    • python model.py. Trains the proposed model on the prepared data and dumps the best models in 'models/'
    • python getRepresent.py. Extracts the representation for each instance so that we can perform oversampling.
    • python sample.py. Use SMOTE to sample theextracted representations.
    • python model-classify.py. Train the classification layers using the sampled data and dumps thebest models in 'models/'.
  • Persona Type Identification
    • python dataloader-model2.py.py. It prepares and dumps the dataframes in a format which can be used by models and dataloaders to train and evaluate.
    • Set appropriate parameters in init_parameter.py
    • sh scripts/run.sh. Trains the proposed model on the prepared data and dumps the best models in 'models/'
  • Persona Value Generation
    • python main.py. Trains the proposed model.