Journal of Frontiers of Computer Science and Technology ›› 2020, Vol. 14 ›› Issue (6): 939-946.DOI: 10.3778/j.issn.1673-9418.1905089

Previous Articles     Next Articles

Asymmetric Recommendation Algorithm in Heterogeneous Information Network

ZHAO Chuan, ZHANG Kaihan, LIANG Jiye   

  1. 1. School of Computer and Information Technology, Shanxi University, Taiyuan 030006, China
    2. Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education, Shanxi University, Taiyuan 030006, China
    3. Institute of Intelligent Information Processing, Shanxi University, Taiyuan 030006, China
  • Online:2020-06-01 Published:2020-06-04

非对称的异质信息网络推荐算法

赵传张凯涵梁吉业   

  1. 1. 山西大学 计算机与信息技术学院,太原 030006
    2. 山西大学 计算智能与中文信息处理教育部重点实验室,太原 030006
    3. 山西大学 智能信息处理研究所,太原 030006

Abstract:

In recent years, as a new research direction, heterogeneous information networks have attracted a lot of attention in recommendation system. Most similarity measures based on heterogeneous information networks follow that the relationship between users is symmetrical, but in real life, due to the number of items rated by different users are various, it sometimes makes the relationship asymmetrical. In order to measure the relationship more accurately between users, first of all, the asymmetric coefficient is employed to characterize the asymmetry of similarity based on the mean square deviation formula. Then according to the characteristics of meta-paths, different weights of meta-paths are given, and the similarity results are weighted to improve accuracy. Finally, the recommendation for rating prediction based on heterogeneous information network is implemented combined with similarity information and rating information in matrix factorization model. The experimental results on dataset demonstrate that the proposed algorithm is superior to the traditional algorithms in the evaluation of mean absolute error and root mean square error.

Key words: heterogeneous information network, meta-path, similarity, matrix factorization

摘要:

异质信息网络作为一个新的研究方向,近年来在推荐系统领域引起了广泛的关注。目前基于异质信息网络的大部分相似度计算方法认为用户的相似关系是对称的,但是在实际中由于不同用户评分的物品数量不同,有时会导致相似关系出现非对称情况。为了能够更好地度量用户之间的相似关系,首先在均方差相似度公式的基础上,引入非对称系数刻画相似度的非对称性;然后根据元路径的特征赋予不同元路径权重,并将不同元路径的相似度结果进行加权以提高用户相似度的准确性;最后通过在矩阵分解模型中融合相似度信息与评分信息实现基于异质信息网络的评分预测推荐。在数据集上的实验结果表明,该算法在平均绝对误差和均方根误差两个评价指标上优于传统算法。

关键词: 异质信息网络, 元路径, 相似度, 矩阵分解