系统管理学报

Journal of Systems & Management ›› 2020, Vol. 29 ›› Issue (1): 61-72.DOI: 10.3969/j.issn.1005-2542.2020.01.007

Previous Articles     Next Articles

Diversity Recommendation Based on Product Co-Purchase Network

WANG Qian, YUJijun   

  1. Business School, Sun Yat-Sen University, Guangzhou 510275, China
  • Online:2020-01-29 Published:2020-05-14

基于商品购买关系网络的多样性推荐

王茜，喻继军   

  1. 中山大学 管理学院，广州 510275
  • 通讯作者: 喻继军( 1984-)，男，博士生。
  • 作者简介:王茜( 1971-)，女，教授，博士生导师。研究方向为电子商务、管理信息系统及决策支持等
  • 基金资助:

    国家自然科学基金资助项目（71772187，70971141，71832014）；

    教育部人文社会科学研究规划资助项目(15YJA630070）；

    广东省自然科学基金资助项目（2014A030313184）

PDF

Abstract: How to provide consumers with diversified product recommendation has become a hot topic in the field of personalized recommendation. However, the traditional diversity recommendation selects diversity products by using the random or reverse scoring method, which is not able to accurately recommend diversity products for consumers. Therefore, by referring the theory of network information diffusion, the purchase record of the consumer is transformed into a commodity purchase relationship network, and the importance of commodity nodes in the network is analyzed by using the full neighborhood method. Using Bayesian association rule algorithm to calculate the key nodes that are related to the products with the highest score in the recommendation prediction, and taking the key nodes as the basis for the recommendation of diversified products, a diversity recommendation method based on the purchase relationship network is proposed. The results show that, compared with the traditional diversity recommendation method, the new recommendation method can more accurately recommend diversified products for consumers, and at the same time, the method can effectively promote the recommendation of long-tail products through the product recommendation diffusion effect formed by the recommendation cascade relationship between key nodes.

Key words: co-purchase network, diversity recommendation, key node, long tail

摘要： 如何为消费者提供多样性商品推荐，已成为个性化推荐领域研究的热点。传统多样性推荐采用随机或评分逆序的方法选择多样性商品，存在无法为消费者准确推荐多样性商品的不足。针对于此，借鉴网络信息扩散的原理，将消费者购买记录二部图转换为商品购买关系网络，采用全邻域方法分析网络中商品节点的重要性；在此基础上，利用贝叶斯关联规则算法计算与推荐预测评分最高商品具有关联关系的关键节点，以此关键节点作为多样性商品推荐的依据，提出基于购买关系网络的多样性推荐方法。研究结果表明，与传统多样性推荐方法相比，新的推荐方法可为消费者更准确地推荐多样性商品的同时，该方法通过关键节点间的推荐级联关系所形成的商品推荐扩散效应，可有效提升长尾商品的推荐。

关键词: 商品购买关系网络, 多样性推荐, 关键节点, 长尾商品

CLC Number: