lancelot39 / mvs

The implementation code of the TOIS paper MVS "Enhancing Multi-View Smoothness for Sequential Recommendation Models"

Enhancing Multi-View Smoothness for Sequential Recommendation Models

This repository contains the code for our TOIS paper Enhancing Multi-View Smoothness for Sequential Recommendation Models.

Overview

We propose MVS, a general training framework for enhancing the multi-view smoothness of sequential recommendation models using implicit feedback data. In our approach, we leverage a complementary model and devise three regularizers to guarantee label smoothness, context smoothness and model smoothness. In this project, we mainly release the implemented code on the SASRec model.

Train MVS

In the following section, we describe how to train a SASRec+MVS model by using our code.

Environment

To faithfully reproduce our results, please follow the same environment as in our previous project S3-Rec

Data

We also utilize the released data from S3-Rec

Training scripts

In our approach, we first train a complementary model (e.g., SASRec), and then utilize it to help train our target model.

cd code
python run_base_sample.py --data_name Beauty
python run_mvs_sample.py --data_name Beauty

We also provide the code under the full-sort ranking setting:

cd code
python run_base_full.py --data_name Beauty
python run_mvs_full.py --data_name Beauty

Citation

Please cite our paper if you use MVS in your work:

@article{zhou2023mvs,
   title={Enhancing Multi-View Smoothness for Sequential Recommendation Models},
   author={Zhou, Kun and Wang, Hui and Wen, Ji-Rong and Zhao, Xin},
   booktitle = {{TOIS}},
   year={2023}
}