A Distributed Recommender System Based on Graph Partition

doi:10.11871/jfdc.issn.2096-742X.2024.05.010

Abstract

Abstract:

[Objective] It is of great significance to design a recommender system with high data processing efficiency. [Methods] The graph structure is used to simulate the user preference relationship in the recommender system. Through the graph partition algorithm processing, the information value of the data in the recommender system can be further mined, and the obtained subgraph data with load balancing can be used as the input of the distributed system. Finally, a distributed recommender system is implemented through the fusion of an adaptive aggregation module. [Results] The system can improve the processing efficiency of the recommender algorithm for large-scale data. On the premise that the prediction accuracy does not decline, the algorithm can improve the efficiency 6.4 times in a cluster training consisting of 16 CPUs compared with a single CPU training. [Conclusions] The experimental results show that the system is effective in recommendation efficiency.

Key words: recommender system, graph partition, load balancing, distributed system

YANG Jinguang, XIONG Fei, GU Junyu, XI Weiting. A Distributed Recommender System Based on Graph Partition[J]. Frontiers of Data and Computing, 2024, 6(5): 102-110, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2024.05.010.

Figures/Tables 13

Fig.1

Table 1

Fig.2

Fig.3

Table 2

Table 3

Table 4

Fig.4

Fig.5

Fig.6

Fig.7

Table 5

Table 6

References 17

[1]	GRECHANIK M, FU C, XIE Q, et al. A search engine for finding highly relevant applications[C]// Acm/ieee International Conference on Software Engineering. ACM, 2010: 475-484.
[2]	JIE L, DIANSHUANG W, MINGSONG M Z, et al. Recommender system application developments: A survey[J]. Decision Support Systems, 2015, 74: 12-32.
[3]	FUYU L, TAIWEI J, CHANGLONG Y, et al. SDM: Sequential Deep Matching Model for Online Large-scale Recommender System[C]// Conference on Information and Knowledge Management, 2019: 2635-2643.
[4]	COVINGTON P, ADAMS J, SARGIN E. Deep Neural Networks for YouTube Recommendations[C]// Acm Conference on Recommender Systems. ACM, 2016: 191-198.
[5]	SHUMPEI O, YUKIHIRO T, SHINGO O, et al. Embedding-based News Recommendation for Millions of Users[C]// Knowledge Discovery and Data Mining, 2017: 1933-1942.
[6]	MCLAUGHLIN M R, HERLOCKER J L. A collaborative filtering algorithm and evaluation metric that accurately model the user experience[C]// International Acm Sigir Conference on Research & Development in Information Retrieval. ACM, 2004: 329-336.
[7]	KOREN Y, BELL R, VOLINSKY C. Matrix Factorization Techniques for Recommender Systems[J]. IEEE, 2009, 42(8): 30-37.
[8]	HE X, DENG K, WANG X, et al. LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation[C]// International ACM SIGIR Conference on Research and Development in Information Retrieval, 2020: 639-648.
[9]	CHEN M, LIHENG M, YINGXUE Z, et al. Memory Augmented Graph Neural Networks For Sequential Recommendation[C]// National Conference on Artificial Intelligence, 2020, 34: 5045-5052.
[10]	DONG H K, CHANYOUNG P, JINOH O, et al. Convolutional Matrix Factorization for Document Context-Aware Recommendation[C]// Conference on Recommender Systems, 2016: 233-240.
[11]	STEFFEN R, CHRISTOPH F, ZENO G, et al. BPR: Bayesian personalized ranking from implicit feedback[C]// Uncertainty in Artificial Intelligence, 2012: 452-461.
[12]	ZHANG W, WANG J Y, FENG W. Combining latent factor model with location features for event-based group recommendation[M]. Knowledge Discovery and Data Mining, 2013: 910-918.
[13]	GUO Q Y, ZHUANG F Z, QIN C, et al. A Survey on Knowledge Graph-Based Recommender Systems[J]. IEEE Annals of the History of Computing, 2022, 34(8): 3549-3568.
[14]	WU Z H, PAN S R, CHEN F W, et al. A Comprehensive Survey on Graph Neural Networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(1): 4-24.
[15]	GEORGE K, VIPIN K. A Fast and High Quality Multilevel Scheme for Partitioning Irregular Graphs[J]. SIAM Journal on Scientific Computing, 1999, 20(1): 359-392.
[16]	MAXWELL H F, JOSEPH A K. The MovieLens Datasets: History and Context[J]. ACM transactions on interactive intelligent systems, 2016, 5(4): 1-19.
[17]	JOHN C D, ELAD H, YORAM S. Adaptive Subgradient Methods for Online Learning and Stochastic Optimization[J]. Journal of Machine Learning Research, 2011, 12(61): 2121-2159.

符号	描述
U	用户集合
V	项目集合
Y	用户-项目交互
N	分区数量
Pⁱ	第i个分区的用户嵌入表示
Qⁱ	第i个分区的项目嵌入表示
Pⁱ_t	转移到相同特征空间上第i个分区的用户嵌入表示
Qⁱ_t	转移到相同特征空间上第i个分区的项目嵌入表示
P	聚合后的用户嵌入表示
Q	聚合后的项目嵌入表示

数据集	用户	项目	用户-项目交互	密度
Movielens-1 m	6,940	3,706	1,000,209	3.89%
Movielens-10 m	71,567	10,681	10,000,054	1.31%

分区数N	1	2	4	8	16
recall@10	0.0693	0.0740	0.0757	0.0723	0.0714
recall@20	0.1135	0.1178	0.1282	0.1265	0.1250
precision@10	0.1681	0.1763	0.1803	0.1776	0.1749
precision@20	0.1422	0.1457	0.1550	0.1501	0.1492

分区数N	1	2	4	8	16
recall@10	0.1050	0.1101	0.1180	0.1206	0.1029
recall@20	0.1684	0.1835	0.1899	0.1940	0.1814
precision@10	0.1715	0.1738	0.1826	0.1845	0.1701
precision@20	0.1437	0.1503	0.1550	0.1563	0.1445

	1	2	4	8	16
子模型训练	—	28.3	15	7.8	4.2
聚合训练	—	8.3	8.3	8.3	8.3
总计	80	36.6	23.3	16.1	12.5