基于多源大数据的个性化推荐系统效果研究  

Research on the Effectiveness of Personalized Recommender System Based on Multi-source Big Data

作　　者：姚凯 涂平[2] 陈宇新[3] 苏萌 YAO KaiI;TU Ping;CHEN Yuxin;SU Meng(Business School,Central University of Finance and Economics,Beijing 100081,China;Guanghua School of Management,Peking University,Beijing 100871,China;Business,New York University Shanghai,Shanghai 200122,China;Percent Corporation,Beijing 100101,China)

机构地区：[1]中央财经大学商学院,北京100081 [2]北京大学光华管理学院,北京100871 [3]上海纽约大学商学部,上海200122 [4]北京百分点信息科技有限公司,北京100101

出　　处：《管理科学》2018年第5期3-15,共13页Journal of Management Science

基　　金：国家自然科学基金(71332006;71702208)~~

摘　　要：个性化推荐系统已成为各大电商向消费者提供个性化购物体验的重要工具之一,通过推荐系统,商家可以提高收入和消费者满意度。但传统推荐系统通常只利用消费者在当前网站的历史信息推荐个性化商品,无法获得消费者在其他网站的数据来优化推荐效果。大数据时代,一些第三方公司抓住机遇,利用不同公司的多源大数据提供更好的个性化推荐服务。然而,这种新型的推荐系统对消费者购物行为的影响存在极大的未知性。探究基于多源大数据的个性化推荐系统对消费者购物行为的影响。为了建立推荐系统与消费者购物行为之间的因果关系,采用实地实验有效地避免传统研究方法存在的内生性问题,并具有较好的外部有效性。一方面,基于内部数据和外部数据构造解释性变量,探究内部数据特征和外部数据特征与推荐效果之间的关系;另一方面,通过检验消费者特征与内外部数据的推荐效果间的交互效应,进一步分析外部数据和内部数据的推荐效果如何随消费者的特征变化,帮助企业更好地利用多源大数据提升推荐效果。研究结果表明,基于内部数据的推荐系统能够显著提升消费者点击个性化推荐商品的概率,可以降低消费者决策时间,激励消费者浏览更多的商品。外部数据的推荐效果不仅与外部公司网站的用户数量相关,也会受到外部网站与当前网站的关联程度的影响。消费者特征对基于内部数据和外部数据的推荐效果起调节作用,如果消费者是当前网站的老用户,利用该消费者在当前网站的内部数据提供个性化推荐的效果更佳。通过分析基于多源大数据的推荐效果对消费者购物行为的影响,进一步完善个性化推荐领域的理论框架。研究结果对如何利用多源数据构建更加有效的推荐系统具有重要指导价值,并为不同网站之间的数据共享机制提供重要的�Personalized recommender system has become one of the important tools for online firms to offer personalized service and unique experience to the consumers. Through the personalized recommender system, merchants can improve their perform- ance and consumer＇s satisfaction. The traditional recommender systems only use consumers＇ historical information of the focal firm to suggest personalized products. They cannot access to the extra data from external firms to achieve higher recommendation performance. In the era of big data, some third-party companies seize the opportunity to use the muhi-source big data from differ- ent companies to provide better personalized recommendation service. However, how does this new type of recommender system influences consumers＇ shopping behavior is largely unknown. This study mainly explores the influence of personalized recommender system based on multi-source big data on consumers＇ online shopping behaviors. In order to establish the causal relationship between recommender system and consumers＇ shopping behaviors, this study adopts field experiment to avoid endogenous problem and achieve high external validity. Meanwhile, ac- cording to constructing explanatory variables from internal and external big data, this study illustrates how the characteristics of muhi-source big data influence the recommendation effectiveness. Finally, this research examines the interaction between con- sumers＇ characteristics and performance of recommender system based on internal and external data to explore how the effective- ness of recommender system varies according to consumers＇ characteristics. The results demonstrate that the recommender system based on internal data can significantly improve consumers＇ click through rate on the personalized products. Meanwhile, the recommender system can help consumers decrease the time cost on de- cision making and stimulate them to browse more products. The performance of recommender system based on external data is re- lated to the numbers

关 键 词：多源大数据 个性化推荐 实地实验 电子商务 互联网营销 

分 类 号：F713.36[经济管理—产业经济]