COVID Doctor Chatbot

Create a doctor chatbot that carries out conversation regarding COVID-19

Overview

We present a Seq2Seq based conversational agent (chatbot) that can answer queries from patients regarding topics related to COVID-19.

The model uses patients’ COVID related questions and previous replies of the certified doctors scraped from the internet as input [1]. The model is trained with the attention based Transformer [2] architectures like GPT [3] and BART [4] to ensure it pays attention to important part of the input sequence when decoding.

Sample Output

Short user query	Long user query

See output.zip for all outputs on test data.

Set up the virtual environment

python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txt

Notebooks

Notebooks/Project_Notebook_BART.ipynb for fine-tuning BART model on covid dataset Notebooks/Project_Notebook_GPT2.ipynb for fine-tuning GPT2 model on covid dataset Notebooks/Project_Notebook_Transformer.ipynb for training Transformer model on covid dataset

Run evaluate script

python3 evaluate.py --o output/transformer_output.txt 
python3 evaluate.py --o output/gpt2_output.txt 
python3 evaluate.py --o output/Bart_output.txt 

Misc

Some of the Transformer outputs generated based on previous conversation and print as new samples, so the number of samples would be larger than others. For example:

--------------------example 53--------------------
Patient: is it safe to get care for my other medical conditions during this time ?
Reference: yes , it is important to continue taking care of your health and wellness . continue your medications , and do not change your treatment plan without talking to your healthcare provider . continue to manage your disease the way your healthcare provider has told you .
Predict: hello , your question on healthcare magic . in the of get cause as yes pneumonia .
--------------------example 54--------------------
Patient: is it safe to get care for my other medical conditions during this time ? [SEP] yes , it is important to continue taking care of your health and wellness . continue your medications , and do not change your treatment plan without talking to your healthcare provider . continue to manage your disease the way your healthcare provider has told you . [SEP] thanks a lot doctor .
Reference: no problem , let us know if you come across anything else .
Predict: brief opinion : ? covid - 19 in have from you like to - 19 . yes should if your . your in you like to covid - do with me ?

Acknowledgment

• [1] Covid MedDialog Dataset from UCSD-AI4H
• [2] Transformer adapted from this implementation
• [3] GPT2 implementation adapted from DialoGPT
• [4] BART implementation adapted from this blog

Our Team

• Philip Cho
• Chenhao Wang
• Azmarie Wang