基于SVD++隐语义模型的信任网络推荐算法

doi:10.11959/issn.2096-0271.2021041

摘要/Abstract

摘要：

推荐算法通常基于用户的行为数据进行建模，然而显式行为数据的稀疏性可能会引起推荐算法的冷启动问题。为了降低数据稀疏和冷启动问题对推荐算法效果的影响，在已有显式信任关系的基础上，基于用户相似度引入隐式信任关系，通过SVD++隐语义模型设计了新的推荐算法。为了提升算法效果，进一步融合邻域模型，推导出算法评分预测式及损失函数。在Epinions开源数据集中将RMSE和MAE作为测试指标，在全体用户集和冷启动用户集上进行对比实验。实验结果显示，设计的推荐算法可以在一定程度上改善原推荐算法的冷启动问题，并取得更好的评分预测效果。

关键词: 推荐算法, 隐语义模型, 信任网络, 评分预测

Abstract:

Recommender algorithms are usually modeled based on user behavior data.However, the sparseness of explicit behavior data may cause the cold start problem of recommender algorithms.In order to solve the impact of data sparseness and cold-start problems on the effect of recommender algorithms, implicit trust relationship based on user similarity was introduced based on the existing revealed trust relationship, and a new recommender algorithm was designed through the SVD++ implicit semantic model.In order to improve the effect of the algorithm, the neighborhood model was integrated further, and the algorithm score prediction formula and loss function were derived.In the Epinions open source data set, RMSE and MAE were used as test indicators, and comparative experiments were conducted on the entire user set and the cold start user set.The experimental results show that the recommender algorithm can optimize the cold start problem of the original recommender algorithm to a certain extent, and achieve a better rating prediction accuracy.

Key words: recommender algorithm, latent factor model, trust network, rating prediction

中图分类号:

TP315

陈佩武, 束方兴. 基于SVD++隐语义模型的信任网络推荐算法[J]. 大数据, 2021, 7(4): 105-116.

Peiwu CHEN, Fangxing SHU. A recommender algorithm based on SVD ++model under trust network[J]. Big Data Research, 2021, 7(4): 105-116.

图/表 8

表1

图1

表2

表3

表4

表5

表6

图2

参考文献 22

[1]	NIE F P , WANG X Q , HUANG H . Clustering and projected clustering with adaptive neighbors[C]// Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York:ACM Press, 2014: 977-986.
[2]	WANG X B , LEI Z , GUO X J ,et al. Multview subspace clustering with intactnessaware similarity[J]. Pattern Recognition, 2019,88: 50-63.
[3]	KOREN Y , BELL R , VOLINSKY C . Matrix factorization techniques for recommender systems[J]. Computer, 2009,42(8): 30-37.
[4]	RESNICK P , IACOVOU N , SUCHAK M ,et al. GroupLens:an open architecture for collaborative filtering of netnews[C]// Proceedings of the 1994 ACM Conference on Computer Supported Cooperative Work. New York:ACM Press, 1994: 175-186.
[5]	GOLDBERG D , NICHOLS D , OKI B M ,et al. Using collaborative filtering to weave an information Tapestry[J]. Communications of the ACM, 1992,35(12): 61-70.
[6]	SARWAR B , KARYPIS G , KONSTAN J ,et al. Item-based collaborative filtering recommendation algorithms[C]// Proceedings of the 10th International Conference on World Wide Web. New York:ACM Press, 2001: 285-295.
[7]	YANG B , LEI Y , LIU D Y ,et al. Social collaborative filtering by trust[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017,39(8): 1633-1647.
[8]	LINDEN G , SMITH B , YORK J.Amazon . com recommendations:item-to-item collaborative filtering[J]. IEEE Internet Computing, 2003,7(1): 76-80.
[9]	BREESE J S , HECKERMAN D , KADIE C . Empirical analysis of predictive algorithms for collaborative filtering[C]// Proceedings of the 14th Conference on Uncertainty in Artificial Intelligence. New York:ACM Press, 1998: 43-52.
[10]	GOLDBERG K , ROEDER T , GUPTA D ,et al. Eigentaste:a constant time collaborative filtering algorithm[J]. Information Retrieval, 2001,4(2): 133-151.
[11]	BLEI D M , NG A Y , JORDAN M I . Latent dirichlet allocation[J]. The Journal of Machine Learning Research, 2003,3: 993-1022.
[12]	KOREN Y . Factor in the neighbors:scalable and accurate collaborative filtering[J]. ACM Transactions on Knowledge Discovery from Data, 2010,4(1).
[13]	XIANG L , YUAN Q , ZHAO S W ,et al. Temporal recommendation on graphs via long-and short-term preference fusion[C]// Proceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York:ACM Press, 2010: 723-732.
[14]	贾俊, 张斌, 李志远 . 基于用户行为分析的个性化推荐算法[J]. 智能科学与技术学报, 2019,1(4): 421-426.
	JIA J , ZHANG B , LI Z Y . Personalized recommendation algorithm based on user behavior analysis[J]. Chinese Journal of Intelligent Science and Technology, 2019,1(4): 421-426.
[15]	AI J , LIU Y Y , SU Z ,et al. Link prediction in recommender systems based on multifactor network modeling and community detection[J]. Europhysics Letters, 2019,126(3): 38003.
[16]	XIONG F , WANG X M , CHENG J J . Subtle role of latency for information diffusion in online social networks[J]. Chinese Physics B, 2016,25(10): 108904.
[17]	JAMALI M , ESTER M . A matrix factorization technique with trust propagation for recommendation in social networks[C]// Proceedings of the 4th ACM Conference on Recommender Systems. New York:ACM Press, 2010: 1055-1066.
[18]	MA H , YANG H X , LYU M R ,et al. SoRec:social recommendation using probabilistic matrix factorization[C]// Proceedings of the 17th ACM Conference on Information and Knowledge Management. New York:ACM Press, 2008: 931-940.
[19]	XU M H , LIU S H . Semantic-enhanced and context-aware hybrid collaborative filtering for event recommendation in event-based social networks[J]. IEEE Access, 2019,7: 17493-17502.
[20]	YANG B , ZHAO P F , PING S Q ,et al. Improving the recommendation of collaborative filtering by fusing trust network[C]// Proceedings of the 8th International Conference on Computational Intelligence and Security. Piscataway:IEEE Press, 2012: 195-199.
[21]	FANG H , BAO Y , ZHANG J . Leveraging decomposed trust in probabilistic matrix factorization for effective recommendation[C]// Proceedings of the 28th AAAI Conference on Artificial Intelligence. Palo Alto:AAAI Press, 2014.
[22]	LIU Y , YANG C , MA J ,et al. A social recommendation system for academic collaboration in undergraduate research[J]. Expert Systems, 2018,36(1): e12365.

对比项	隐语义模型	邻域模型
理论基础	有监督的机器学习	统计、协同过滤
时间复杂度	O(k · f ·d)	UserCF：O(n· (k/n)²)
		ItemCF：O(m· (k/m)²)
空间复杂度	O((m+n) · f)	UserCF: O(m·m)
		ItemCF: O(n·n)
推荐解释	无法提供推荐原因解释	ItemCF可以解释推荐结果
实时推荐	每次生成用户推荐列表时需要重新计算，不适合实时推荐计算	UserCF和ItemCF可将用户和物品表缓存，从而进行实时预测

数据集	用户数	物品数	评分数	评分数据密度	信任者数	被信任者数	信任关系数	信任数据密度
Epinions	40 163	139 738	664 824	0.011 8%	33 960	49 290	487 181	0.029%

数据集	用户ID范围	物品ID范围	评分值范围	信任者 ID范围	被信任者ID范围信任值范围
Epinions	[1, 49 290]	[1, 139 738]	[1, 5]	[1, 49 290]	[1, 49 290]1 / null

参数	隐语义因子数d	迭代次数iter	学习速率α	λ	λ_t1	λ_t2
值	5 / 10	100	0.01	0.9	0.8	0.5

隐向量维度	测评指标	UserMean	ItemMean	SVD++	TrustMF	本文算法	提高程度
d=5	MAE	1.147	0.902	0.889	0.890	0.870	2.1%
d=5	RMSE	1.430	1.127	1.162	1.109	1.102	2.2%
d=10	MAE	1.147	0.902	0.891	0.853	0.860	-0.8%
d=10	RMSE	1.430	1.127	1.166	1.114	1.102	1.1%