计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (23): 175-179.DOI: 10.3778/j.issn.1002-8331.1908-0428

• 模式识别与人工智能 • 上一篇    下一篇

序列生成对抗网络在推荐系统中的应用

伍鑫,黄勃,方志军,刘文竹   

  1. 1.上海工程技术大学 电子电气工程学院,上海 201620
    2.江西省经济犯罪侦查与防控技术协同创新中心,南昌 330000
  • 出版日期:2020-12-01 发布日期:2020-11-30

Application of Sequence Generative Adversarial Network in Recommendation System

WU Xin, HUANG Bo, FANG Zhijun, LIU Wenzhu   

  1. 1.School of Electrical and Electronic Engineering, Shanghai University of Engineering and Technology, Shanghai 201620, China
    2.Jiangxi Collaborative Innovation Center for Economic Crime Detection and Prevention and Control, Nanchang 330000, China
  • Online:2020-12-01 Published:2020-11-30

摘要:

推荐系统旨在根据用户的历史行为数据发现该用户可能感兴趣的新项目,并产生相应的推荐。当前大部分的推荐系统多根据用户的历史行为数据,挖掘相似用户,并从相似用户的历史数据中选出彼此历史数据中未出现的新项目;或者根据用户感兴趣的历史项目匹配相似的新项目,从而实现推荐。但这些推荐方式对原始数据有着较强的依赖关系,且难以发觉不同项目之间隐含的序列关系。因此提出一种融合Item2vec和生成对抗网络(Generative Adversarial Networks,GAN)方法的推荐算法,可以学习得到项目间难以表达的关系;挖掘用户历史数据中的序列关系,学习用户兴趣偏好的真实分布;实现用户兴趣偏好的预测。实验发现该推荐算法具有较好的表现。

关键词: 推荐系统, 序列预测, Item2vec, 生成对抗网络(GAN), 偏好特征

Abstract:

The recommendation system is designed to discover new items that the user may be interested in based on the user’s historical behavior data and generate corresponding recommendations. Most studies mine similar users according to the user’s historical behavior data, and select items from the historical data of similar users, which is not appeared in the historical data of the target user, or match similar items according to user history to achieve recommendations. However, these recommendation methods have strong dependence on the original data, and it is difficult to detect the implicit sequence relationship between different items. Therefore, a recommendation algorithm which combines the Item2vec and Generative Adversarial Networks(GAN) is proposed in this paper, which can learn the relationship between different items. And this method can find the sequence relationship in the users’ history data to learn the distribution of users’ preference. Furthermore, the users’ preference can be predicted. The experiment is performed and this method has great improvement.

Key words: recommendation system, sequence prediction, Item2vec, Generative Adversarial Networks(GAN), preference feature