RMC - Recommending Missed Citations Identified by Reviewers

This repository contains datasets and code for our COLING 2024 paper: "Recommending Missed Citations Identified by Reviewers: A New Task, Dataset and Baselines".

The CitationR Dataset

CitationR is created by extracting recommended citations in reviews from NeurIPS and ICLR. In total, we collect 76,143 official reviews and 21,598 submissions, among which around 35% of submissions are identified as lacking citations. Moreover, to better replicate the actual situation in which researchers search for papers to cite, we establish a larger and more challenging version of CitationR. This version includes additional 40,810 papers published in top venues that reviewers frequently recommend citations from.

Download the base version from the following link: Baidu Drive (access code: q2vy)

Download the extended version from the following link: Baidu Drive (access code: gjwu)

Setup

Data Preparation

Unzip downloaded data and put them under data/ with names of base and extended respectively.

Environment Setup

conda create -n rmc python==3.7.12
source activate rmc
pip install -r requirements.txt

Baselines

BM25

• Evaluation

python ./src/bm25.py

Citeomatic

Follow official introduction in citeomatic

HTransformer

Follow official introduction in Local-Citation-Recommendation

Pre-trained LM (BERT, SciBERT, SPECTER, CiteBERT, LinkBERT, SciNCL ...)

• Evaluation of LMs not fine-tuned on CitationR

python ./src/evaluating/direct_plm_evaluating.py --dataset_name base --model_name scincl --way_ref concat

• Evaluation of LMs fine-tuned on CitationR

python ./src/evaluating/evaluate.py --dataset_name base --config_name example --experiment_name example_evaluation

• Train

python ./src/training/run_training.py --model scincl --dataset base --config_name example --experiment_suffix example_train

Our fine-tuned models are available: base(access code: nrbe), extended(access code: n3pb).

Put downloaded models under experiments/dataset_name/example_evaluation/model/

Citation

If you find our work useful, please cite the paper as:

@inproceedings{long24coling,
  title = {Recommending Missed Citations Identified by Reviewers: A New Task, Dataset and Baselines},
  author = {Kehan Long and Shasha Li and Pancheng Wang and Chenlong Bao and Jintao Tang and Ting Wang},
  booktitle = {COLING},
  year = {2024}
}

Acknowledgement

• We use some of the code in Local-Citation-Recommendation and specter for implementing our project.