Exploring Beyond-Accuracy in ChatGPT’s Recommendations

This repository contains the source code and datasets for the paper titled "Exploring Diversity, Novelty, and Popularity Bias in ChatGPT’s Recommendations".

To correctly clone the repo, use:

git clone --recursive XYZ

The project includes the necessary source code and files to conduct experiments for the baselines, ChatGPT, GPT-3.5 and PaLM-2. Before running these codes, ensure that you have Python version 3.8.6 or later installed on your device. To set up the required environment, you can create a virtual environment and install the necessary dependencies using the provided requirements files with the following steps:

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

Alternatively, you can use conda to recreate the same environment by executing the following command:

conda env create -f env/env_recommender-ChatGPT.yml
conda activate Recommender-ChatGPT

To correctly initialize Elliot, from ../Recommender-ChatGPT/external/:

pip install --upgrade pip
pip install -e . --verbose

To obtain the results, after inserting the API token, execute the following command:

python code/main.py

The results are stored in the results folder.