Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (13): 102-111.DOI: 10.3778/j.issn.1002-8331.2108-0328

• Big Data and Cloud Computing • Previous Articles     Next Articles

Hybrid Recommendation Algorithm Combining Multi-Semantic Trust and Global Knowledge

WANG Yonggui, CAI Yongwang, WANG Yang   

  1. School of Electronic and Information Engineering, Liaoning Technical University, Huludao, Liaoning 125105, China
  • Online:2022-07-01 Published:2022-07-01

融合多语义信任度与全局信息的混合推荐算法

王永贵,蔡永旺,王阳   

  1. 辽宁工程技术大学 电子与信息工程学院,辽宁 葫芦岛 125105

Abstract: The problem of data sparsity generally exists in collaborative filtering system. The similarity measurement which only considers the local context information on the common rating items does not have high reliability. In order to solve the above problems, a hybrid recommendation algorithm MSTGK is proposed, which integrates multi-semantic trust and global knowledge. Firstly, the weighted heterogeneous information network(WHIN) is introduced to deal with the impact of rating data, social relations, user tags and item attributes on user trust through weighted meta path, and mine trust information with different semantics to alleviate the problem of data sparsity. Secondly, considering the influence of popularity of items and user preference on user similarity, they are used as weight factors to improve the JMSD similarity measure in order to improve the accuracy of similarity calculation. Finally, the user’s multi-semantic trust and global similarity are integrated for comprehensive recommendation. Experimental results on two real datasets, DoubanMovie and Yelp, show that the proposed method alleviates data sparsity and compared with other baseline methods, the prediction accuracy is improved by 2.01 and 2.45?percentage points respectively.

Key words: collaborative filtering, weighted heterogeneous information network(WHIN), weighted meta path, trust relationship, project popularity

摘要: 数据稀疏问题普遍存在于协同过滤系统,仅考虑共同评分项目上局部上下文信息的相似度度量方法已不具备较高可靠性。为解决上述问题,提出一种融合多语义信任度和全局信息的混合推荐算法(multi semantic trust and global knowledge,MSTGK)。引入加权异构信息网络(weighted heterogeneous information network,WHIN),通过加权元路径处理评分数据、社交关系、用户标签和项目属性对用户信任的影响,挖掘不同语义的信任信息以缓解数据稀疏性问题;考虑项目流行度和用户偏好程度两个全局要素对用户相似度的影响,将其作为权重因子改进了JMSD相似测度,旨在提高相似度计算精度;融合用户的多语义信任度和全局相似度进行综合推荐。在DoubanMovie和Yelp两个真实数据集上的实验结果表明,所提算法缓解了数据稀疏问题,相比于其他基线方法,预测准确率分别提高了2.01个百分点和2.45个百分点。

关键词: 协同过滤, 加权异构信息网络(WHIN), 加权元路径, 信任关系, 项目流行度