计算机与现代化 ›› 2020, Vol. 0 ›› Issue (12): 13-19.

• 网络与通信 • 上一篇    下一篇

基于异构信息网络的推荐系统

  

  1. (青岛科技大学信息科学技术学院,山东青岛266100)
  • 出版日期:2021-01-07 发布日期:2021-01-07
  • 作者简介:崔鑫(1994—),男,山东淄博人,硕士研究生,研究方向:数据挖掘,推荐系统,E-mail: 1097738469@qq.com。
  • 基金资助:
    国家自然科学基金资助项目(61973180)

Recommendation System Based on Heterogeneous Information Network

  1. (College of Information Science and Technology, Qingdao University of Science and Technology, Qingdao 266100, China)
  • Online:2021-01-07 Published:2021-01-07

摘要: 随着互联网、计算机等技术的深入发展,互联网为用户带来了各类网络服务用于增进用户交流。其中,问答社区为用户提供了提问和回答的交流平台,其目的是通过互联网实现用户间的知识经验分享和信息传播。但仍存在一些问题限制问答社区的发展,例如随着用户数量的不断增长,大量问题得不到及时回答且提问者对已有问题的回答并不满意。因此,对于问答社区来说,如何从大量的用户中找到专家用户是非常重要的。针对以上问题,本文提出一种基于异构信息网络的推荐方法,首先对问答社区中的问题属性和用户属性建立异构信息网络,利用元路径来捕捉异构信息网络中丰富的语义信息,然后使用基于元路径的相似度计算方法分别计算问题与用户的相似度矩阵,采用3种方式将得到的相似度矩阵与问题-用户评分矩阵相融合,然后使用矩阵分解获得问题和用户的潜在特征,最后使用因子分解机进行训练和推荐。在海川化工问答数据集上将本文提出的方法同多种先进的推荐算法进行对比,并利用评价指标对模型进行评估。实验结果表明,本文提出的算法在相关评估指标方面相较于之前的算法具有一定优势。

关键词: 异构信息网络, 问答社区, 协同过滤, 因子分解机

Abstract: With the development of Internet, computer and other technologies, Internet has brought various network services for users to enhance communication among users. Among them, the community question answering provides users with a communication platform for questions and answers, the purpose of which is to achieve knowledge and experience sharing and information dissemination among users through Internet. However, there are still some problems that limit the development of the community question answering. For example, as the number of users continues to increase, a large number of questions cannot be answered in time and the questioners are not satisfied with the answers to existing questions. Therefore, for the question and answer community, how to find experts from a large number of users is very important. In response to the above problems, this paper proposed a recommendation method based on heterogeneous information network. Firstly, a heterogeneous information network was established for the question and user attributes in the question and answer community, and meta-paths were used to capture the rich semantic information in the heterogeneous information network. Secondly, the similarity calculation method based on the meta paths was used to calculate the similarity matrix between the question and the user, and three methods were used to fuse the obtained similarity matrix with the question-user scoring matrix. In the end, the matrix decomposition was used to obtain the potential features of the question and the user. Factorization machine is used for training and recommendation. The method proposed in this paper was compared with various advanced recommendation algorithms on Haichuan chemical community question answering data set, and the evaluation index was used to evaluate the model. Experimental results show that the algorithm proposed in this paper has certain advantages over the previous algorithms in terms of relevant evaluation indicators.

Key words: heterogeneous information network, community question answering, collaborative filtering, factorization machine