Towards Persona-Based Empathetic Conversational Models (PEC)

This is the repo for our work "Towards Persona-Based Empathetic Conversational Models" (EMNLP 2020). The code depends on PyTorch (>=v1.0) and transformers (>=v2.3).

Data

The PEC dataset is available here.

The persona dataset with 100 persona sentences each is available here

You can refer to the sample files here to preprocess the datasets: valid_cleaned_bert.pkl and persona_20.pkl

Our dataset is available on Huggingface Datasets for ease of use.

Model

This repo includes our implementation of CoBERT.

Training

python CoBERT.py --config CoBERT_config.json

Evaluation

Set test_mode=1 and load_model_path to a saved model in CoBERT_config.json, and then run

python CoBERT.py --config CoBERT_config.json