This repo contains a collection of scripts to experiment with NLP understanding of behavioral messages.
Currently working with BCT Taxonomy v1.
A dataset of behavioral messages is generated from a prompt file in ./prompts/ directory.
A prompt file either
- consists of two lines: system prompt and user prompt
- or contain
=====line, in which case everything above this line is the system prompt and everything below is the user prompt
Use a prompt file prompt_name.txt to generate a dataset with num messages per BCT via the command
python -m create_dataset -p prompt_name -n num
The default values for the arguments are prompt_name=baseline and num=10.
A created dataset will be saved in ./data/prompt_name/ as a series of .csv files named after each of the BCT numbers.