IntEL: Intent-aware ranking Ensemble Learning

Introduction

This repository contains the offical implementation of our SIGIR 2023 paper:

Jiayu Li, et al. "IntEL-Intent-aware Ranking Ensemble for Personalized Recommendation". SIGIR 2023.

In this paper, we propose a novel ranking ensemble method IntEL for intent-aware single-objective ranking lists aggregation. To our knowledge, we are the first to generalize ranking ensemble learning with item-level weights on heterogeneous item lists. And we are also the first to integrate user intents into rank aggregation in recommendation.

Requirements

python 3.6
torch==1.7.1
pygad==2.19.2
numpy, pandas, scikit_learn, scipy, tqdm

All will be installed by requirements.txt

Quick Start

The implementation configs of all baseline models and our model IntEL are shown in ./IntEL/script/.

cd ./IntEL/script/
bash IntEL.sh
bash baselines.sh

You can use self-defined configs and dataset by running ./IntEL/src/main.py.

python main.py --model_name IntEL --datapath ${YOUR_DATASET_PATH} --dataname ${YOUR_DATASET_NAME}

Repository Structure

Data Preparation

Dataset/
- It includes the data format, preprocessing, and basic model training process of Tmall dataset.

Model Implementation

IntEL/
- IntEL/data/: includes a toy dataset used for ranking ensemble learning, which is sampled from Tmall dataset (The whole dataset is omitted since it's too large). To train a model with IntEL, session-level interactions, user and item metadata, and groud truth of intents are necessary.
- IntEL/src/: implementation codes of IntEL and all baseline models. Names of item/user/context features and basic lists need re-definition in helpers/BaseReader.py and helpers/SeqReader.py when applying to new datasets.
- IntEL/script/: includes running configs of all methods that were reported in the paper.

Appendix

Appendix/
- Appendix/Appendix-Proof.pdf: Proof of Theorem 2 (BPR loss) and Theorem 3 (P-L loss) in the paper.

Results

Results of Tmall and LifeData with baseline models and IntEL (by NDCG@3):

Please refer to our paper for more results and in-depth analysis of the results.

Citation

If you find our work useful, please do not save your star and cite our work:

@article{li2023intent,
  title={Intent-aware Ranking Ensemble for Personalized Recommendation},
  author={Li, Jiayu and Sun, Peijie and Wang, Zhefan and Ma, Weizhi and Li, Yangkun and Zhang, Min and Feng, Zhoutian and Xue, Daiyue},
  journal={arXiv preprint arXiv:2304.07450},
  year={2023}
}

JiayuLi-997 / IntEL-SIGIR2023