MoEL: Mixture of Empathetic Listeners

This repository contains the MindSpore implementation of the paper:

MoEL: Mixture of Empathetic Listeners. Zhaojiang Lin, Andrea Madotto, Jamin Shin, Peng Xu, Pascale Fung. EMNLP 2019. PDF.

And this code refers to the PyTorch implementation from paper authors.

Setup

1. Install the required libraries (Python 3.7.5 | CUDA 11.1)

2. Run the command: pip install mindspore-cuda11-dev -i https://pypi.tuna.tsinghua.edu.cn/simple

3. Download Pretrained GloVe Embeddings (glove.6B.300d.txt) and save it in /vectors.

4. Download dataset from https://github.com/HLTCHKUST/MoEL/tree/master/empathetic-dialogue and save it in /empathetic-dialogue.

Training & Testing

python3 main.py --emb_dim 300 --hidden_dim 300 --hop 1 --heads 2 --topk 5 --cuda --pretrain_emb --basic_learner --schedule 10000

Predicted Value

BLEU: 2.5