LogiCoL

This repository contains the datasets and source code used in our EMNLP 2025 accepted paper "LogiCoL: Logically-Informed Contrastive Learning for Set-based Dense Retrieval".

Our data are available at this link.

Installation

This project is implemented using pytorch_lightning. Please ensure you have Python 3.8 or higher installed.

To install all required packages, run the following command:

pip install -r requirements.txt

Training

To train a dense retriever using LogiCoL, run the following command:

./script/train_quest_LogiCoL.sh

To train a dense retriever using the basic supervised constrastive learning (the SupCon baseline), run the following command:

./script/train_quest_SupCon.sh

Inference

To inference using the trained retriever, run the following command:

./script/evaluate_quest.sh

To inference using BM25, run the following command:

./script/evaluate_bm25.sh