Code for a paper "PSYDIAL: Personality-based Synthetic Dialogue Generation using Large Language Models" accpeted in LREC-COLING 2024 Main
- STEP 1 - Personality Setting
- STEP 2 - Profile Selecting
- STEP 3 - Dialogue Generation
- STEP 4 - Dialogue Filtering
- STEP 5 - Dialogue Regeneration
5_personality.csv
This file contains descriptions from 5 dimensions of Big five personality. It is utilized to set the personality of two speakers inSTEP 1 (Personality Setting).personaChat_personality_origin.csv
This file contains persona information from PersonaChat dataset. It is utilized inSTEP 2 (Profile Selecting).
❗ You need an Open AI API key to execute the pipeline. ❗
main_pipeline.py
The main pipeline through STEP 1 to a part of STEP 4 is implemented in this file.main_filtering.py
Part of STEP 4 (Dialogue Filtering) is implemented in this file. It filters generated dialogues.main_regeneration.py
This file is about STEP 5 (Dialogue Regeneration). It regenerates a dialogue filtered as a negative dilaogue.
@misc{han2024psydial,
title={PSYDIAL: Personality-based Synthetic Dialogue Generation using Large Language Models},
author={Ji-Eun Han and Jun-Seok Koh and Hyeon-Tae Seo and Du-Seong Chang and Kyung-Ah Sohn},
year={2024},
eprint={2404.00930},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
