End-to-End Knowledge-Routed Relational Dialogue System for Automatic Diagnosis

By Lin Xu, Qixian Zhou, Ke Gong, Xiaodan Liang, Liang Lin (AAAI19) This is a simple code of KR-DQN for medical diagnosis diaglogue system. The model is trained and evaluated on MuZhi dataset

Requirements

Python3, pytorch 0.4.1

Training

# Training on MZ dataset. 
./scripts/train.sh

The checkpoint models are saved at ./checkpoints/exp_models. Tensorboard logs are saved at ./runs

Inference and Evaluation

# Inference and Evaluation
./scripts/predict.sh

The visible dialogue results and quantitative matrics are showed on terminal.

Reference

@inproceedings{xu2018,
    Author = {Lin Xu, Qixian Zhou, Ke Gong, Xiaodan Liang, Liang Lin},
    Title = {End-to-End Knowledge-Routed Relational Dialogue System for Automatic Diagnosis},
    Booktitle = {AAAI},
    Year = {2019}
}