摘要
用户兴趣模型是电子商务个性化推荐服务的基础,用户兴趣数据的获取则是构建用户兴趣模型的核心环节。传统的轮询采集方法存在数据源不够全面、在线应用可扩展性差的不足,会导致同一业务分析所需数据采集时间跨度大、先采集数据可能失效等情况,使得最终业务分析结果出现偏差。针对上述问题,对B2C电子商务用户兴趣数据进行了深入分析,提出了一种基于智能Agent的多源用户兴趣数据采集机制DFM-IA(Data Fetching Mechanism based on Intelligent Agent)。DFM-IA以用户Session为基本处理单元,设计了四种智能Agent(Fetching Agent、Watching Agent、Sort Agent、Logical Agent)和三条排序规则,对七类用户兴趣数据(浏览行为、关键词搜索、收藏行为、购物车行为、订单行为、支付行为、评价行为)进行排序与合并处理,从而在丰富数据采集源的同时大幅提高了在线数据采集效率,有助于解决推荐服务的数据稀疏性问题。仿真实验表明了该机制的高效性。
How to realize accurate, fast and personalized recommendation service is an important research topic of social computing. Recommender systems can help users improve the quality and efficiency of decision-making, and is considered to be the most effective tool to solve the current problem of information overload. The essence of recommender systems is a user-oriented decision support system, and can be used to help E-commerce enterprise managers run target selection. Thus, recommender systems have been regarded as important strategic considerations for enterprises. User interest model is the fundamental of E-commerce recommender systems. Data collection of user interest is the core work for building user interest model. The traditional polling method has some disadvantages, such as incomprehensive data sources and poor scalability for online applications. These disadvantages will lead to some bad situations. For instance, data collection for analysis of same business may span a big time interval, or prior data may be invalidated. All these bad situations could eventually result in skewed analysis. Hence, B2C websites need excellent data collection methods to provide strong support for personalized recommendation service, which is one type of big data analysis. In Section 1 we introduced and analyzed the related work. These interest data, which are created during the course of user-website interaction, can be divided into two types： explicit ratings and implicit ratings. We analyzed the advantages and disadvantages of the two types of data in detail. In Section 2 we deeply analyzed seven types of user interest data in B2C E-commerce. By the analysis of user-system interaction behaviors in some big B2C E-commerce websites at home and abroad, we drew a graph to describe the interest expression levels of user behaviors. After that, we summarized and analyzed the seven types of user interest data sources, including browsing behavior, keyword search, collect behavior, shopping cart behavior, order behavior, payment behavior, and evaluation behavior. In Section 3, we proposed a multi-source user interest data fetching mechanism based on intelligent agent, i.e. DFM-IA. We introduced the framework and workflow of DFM-IA in detail. DFM-IA treats user session as the basic process unit. DFM-IA incorporated four kinds of intelligent agents, including Fetching Agent, Watching Agent, Sort Agent and Logical Agent, and three sort rules. DFM-IA can use them to sort and merge the above seven types of user interest data. Thus, data sources will be comprehensive, and the efficiency of online data collection is dramatically optimized. It can be helpful for solving the problem of data sparsity when running recommendation service. Moreover, we analyzed theoretically service efficiency and access efficiency of DFM-IA. Simulation experiments were carried out to test DFM-IA in Section 4. Based on our developed simulation experiments platform developed, we finished three experiments by considering various numbers of users, sessions, and data of single session. The experimental results showed the high efficiency of DFM-IA. Section 5 concluded this article. DFM-IA is more suitable for the demands of online applications in B2C websites than the traditional method. Next research will focus on how to design information fusion methods for multi-source user interest data. We could improve the accuracy of recommendation and robustness of information.
作者
李聪
马丽
梁昌勇
LI Cong MA Li LIANG Chang-yong(College of Computer Science, Sichuan Normal University, Chengdu 610068, China Katz Graduate School of Business, University of Pittsburgh, Pittsburgh 15213, USA Library, Sichuan Normal University, Chengdu 610068, China School of Management, Hefei University of Technology, Hefei 230009, China)
出处
《管理工程学报》
CSSCI
CSCD
北大核心
2017年第1期58-70,共13页
Journal of Industrial Engineering and Engineering Management
基金
国家自然科学基金资助项目(71202165)
国家自然科学基金重点资助项目(71331002)
四川省哲学社会科学规划资助项目(SC13C019)
