This repository contains the code and data shared under Academic Free License
for our COLING 2025 paper:
PIRsuader: A Persuasive Chatbot for Mitigating Psychological Insulin
Resistance in Type-2 Diabetic Patients
A demo of PIRsuader setup on NAO robot by Mouad Hakam is here
and a short video presentation of the paper is here.
Setup: The shared code runs successfully with the following library versions on Python 3.11.4
openai==0.28.0
nltk==3.8.1
pytorch-transformers==1.2.0
torch==2.0.1
torchvision==0.15.2
trl==0.9.6
sentence-transformers==2.2.2
transformers==4.39.3
wandb==0.15.11
accelerate==0.28.0
peft==0.5.0
The pip list of our environment is included.
How to run: Generally speaking, we provide each "executable" with a configuration file where you need to specify details such as OPENAI key or hugging-face model names. The three main executables will be related to:
- Conversation Generation (with or without predicted dialog acts--convgen directory)
- Dialog act prediction model (initial model--dact directory, further fine-tuning using Reinforcement Learning--RL directory)
- Reward modeling (--CAR and PBR subdirectories in rewards directory)
Please find the details in "README"s for each directory.