Skip to content
/ BMNS Public

Implementation for "Batch-Mix Negative Sampling for Learning Recommendation Retrievers"

License

Notifications You must be signed in to change notification settings

lqz72/BMNS

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Implementation for BMNS

Batch-Mix Negative Sampling for Learning Recommendation Retrievers, CIKM2023

Datasets

We provide Gowalla dataset for reproduction, you can also download other datasets and process them according to our method.

Data Link

Download raw datasets from: https://recbole.io/dataset_list.html

Preprocess

  1. Edit datasets/raw_data/XXX/XXX.yaml
  2. Rewrite&Run datasets/data_preprocess/inter2mtx.py
  3. The data file is saved in datasets/clean_data

Algs

  • --debias 1 : SSL
  • --debias 2 : SSL-Pop
  • --debias 3 : correct-sfx
  • --debias 4 : MNS, Mixed Negative Sampling
  • --debias 5 : BMNS, Batch-Mix Negative Sampling
  • --debias 6 : BIR, Batch Importance Resampling

Quick Start

  • See files run.sh and run_baseline.sh
  • The experimental results are as follows:
Method/Metric NDCG@10 NDCG@20 NDCG@50 Recall@10 Recall@20 Recall@50
SSL 0.1440±5.0 0.1154±3.1 0.1933±5.9 0.1678±4.3 0.2942±7.0 0.2653±4.08
SSL-Pop 0.1597±9.4 0.1201±6.3 0.2060±13.0 0.1709±10.4 0.3027±8.3 0.2640±7.1
correct-sfx 0.1548±8.4 0.1223±4.6 0.2056±7.3 0.1765±5.2 0.3085±9.7 0.2755±8.2
MNS 0.1602±9.5 0.1207±6.6 0.2069±9.7 0.1718±7.2 0.3038±10.4 0.2651±8.6
BIR 0.1607±10.4 0.1207±6.4 0.2069±12.4 0.1715±8.3 0.3039±11.0 0.2650±9.9
BMNS 0.1720±7.4 0.1344±6.1 0.2270±7.6 0.1935±7.7 0.3391±9.9 0.3022±10.8
Impv. 7.06% 9.88% 9.68% 9.64% 9.92% 9.69%

Cite us

@inproceedings{10.1145/3583780.3614789,
author = {Fan, Yongfu and Chen, Jin and Jiang, Yongquan and Lian, Defu and Guo, Fangda and Zheng, Kai},
title = {Batch-Mix Negative Sampling for Learning Recommendation Retrievers},
year = {2023},
isbn = {9798400701245},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3583780.3614789},
doi = {10.1145/3583780.3614789},
booktitle = {Proceedings of the 32nd ACM International Conference on Information and Knowledge Management},
pages = {494–503},
numpages = {10},
keywords = {information retrieval, negative sampling, recommender systems},
location = {<conf-loc>, <city>Birmingham</city>, <country>United Kingdom</country>, </conf-loc>},
series = {CIKM '23}
}

About

Implementation for "Batch-Mix Negative Sampling for Learning Recommendation Retrievers"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published