北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2021, Vol. 44 ›› Issue (3): 106-111.doi: 10.13190/j.jbupt.2020-229

• 研究报告 • 上一篇    下一篇

可扩展的融合多源异构数据的推荐模型

冀振燕1, 吴梦丹1,2, 杨春1, 李俊东1   

  1. 1. 北京交通大学 软件学院, 北京 100044;
    2. 中国科学院 软件研究所, 北京 100190
  • 收稿日期:2020-11-12 出版日期:2021-06-28 发布日期:2021-06-23
  • 作者简介:冀振燕(1972-),女,副教授,博士生导师,E-mail:zhyji@bjtu.edu.cn.
  • 基金资助:
    国家重点研究发展计划项目(2018YFC0809300);国家自然科学基金项目(51935002)

Scalable Recommendation Models Fusing Multi-Source Heterogeneous Data

JI Zhen-yan1, WU Meng-dan1,2, YANG Chun1, LI Jun-dong1   

  1. 1. School of Software Engineering, Beijing Jiaotong University, Beijing 100044, China;
    2. Institute of Software, Chinese Academy of Sciences, Beijing 100190, China
  • Received:2020-11-12 Online:2021-06-28 Published:2021-06-23

摘要: 社交关系在生活中扮演着重要角色,用户通常会受到其好友偏好的影响,更容易选择好友购买过的物品.为了解决推荐系统冷启动问题,对融合社交关系的推荐系统进行了研究,提出了贝叶斯个性化排序评论评分社交模型和可扩展的贝叶斯个性化排序评论评分社交模型,将评分、评论、社交关系等多源异构数据从数据源层面进行了融合,通过用户好友信任度模型将社交关系引入到推荐系统中,用基于段向量的分布式词袋模型处理评论,用全连接神经网络处理评分,用改进的贝叶斯个性化排序模型对排序结果进行优化.实验在Yelp公开数据集上进行了实验,实验结果表明,所提出的2种模型的推荐准确度均优于其他推荐模型.

关键词: 推荐系统, 多源异构数据, 社交关系, 混合模型, 数据融合

Abstract: Social relationship plays an important role in life, and users are often affected by their friends' preferences. It is easier for users to choose items that their friends have purchased. In order to solve the cold start problem of the recommended system, a recommendation system that integrates social relationships is studied, and Bayesian personalized ranking review score social model and scalable Bayesian personalized ranking review score social model are proposed. The proposed fusion recommendation models integrate multi-source heterogeneous data such as scores, reviews, and social relationships from the data source level, introduce social relationships into the recommendation system through the user friend trust model, use the paragraph vector-distributed memory model to process review, use the fully connected neural network to process rating, and use an improved Bayesian personalized ranking model to optimize the ranking results. Experiments are conducted on the Yelp public dataset. It is shown that the recommendation accuracy of the two proposed models are better than other recommendation models.

Key words: recommendation system, multi-source heterogeneous data, social relation, hybrid model, data fusion

中图分类号: