This repository contains the code, datasets, and pre-trained model used in our paper: Chain-of-Factors Paper-Reviewer Matching.
We use one NVIDIA RTX A6000 GPU to run the evaluation code in our experiments. The code is written in Python 3.8. You can install the dependencies as follows.
conda env create --file=environment.yml --name=cof
conda activate cof
./setup.sh
You need to first download the datasets and the pre-trained model. After you unzip the downloaded files, put the folder (i.e., data/ and model/) under the repository main folder ./.
After that, you can run our evaluation script:
./run.sh
Soft/Hard P@5 and P@10 scores will be shown at the end of the terminal output as well as in ./scores.txt.
We use four datasets - NIPS, SciRepEval, SIGIR, and KDD - in our paper. More details about each dataset are as follows.
| Dataset | #Papers | #Reviewers | #Annotated (Paper, Reviewer) Pairs | Conference(s) | Source |
|---|---|---|---|---|---|
| NIPS | 34 | 190 | 393 | NIPS 2006 | Link |
| SciRepEval | 107 | 661 | 1,729 | NIPS 2006, ICIP 2016 | Link |
| SIGIR | 73 | 189 | 13,797 | SIGIR 2007 | Link |
| KDD | 174 | 737 | 3,480 | KDD 2020 | Newly constructed by us |
If you find our code, model, or the KDD dataset useful in your research, please cite the following paper:
@inproceedings{zhang2025chain,
title={Chain-of-factors paper-reviewer matching},
author={Zhang, Yu and Shen, Yanzhen and Kang, SeongKu and Chen, Xiusi and Jin, Bowen and Han, Jiawei},
booktitle={WWW'25},
year={2025}
}