【作者】 张一平；

【导师】 金淳；

【作者基本信息】 大连理工大学 ， 电子商务与物流管理， 2011， 硕士

【摘要】 随着电子商务环境下“信息过载”问题的日益严重,个性化服务推荐作为一种有效的解决途径受到社会各界的普遍重视,成为新兴电子商务研究领域之一。而随着移动网络技术的快速发展,在新兴移动电子商务环境下考虑情境的精确化个性推荐问题成为当前迫切需要解决的一项关键技术问题。个性化推荐的传统研究方法存在数据稀疏、冷启动、数据收集难等致命问题,推荐策略的好坏需要进行考察和评估,移动商务环境下情境因素对顾客消费行为的影响不可忽略,采用基于Agent的建模与仿真方法不仅可以很好地解决上述问题,而且具有捕捉系统宏观“涌现性”的能力。因此,本文采用ABMS方法进行个性化推荐研究,以餐饮推荐系统为例,通过个体交互作用所产生的涌现特性来分析顾客的消费行为以及个性化推荐策略的有效性,论文的核心工作如下：(1)将ABMS方法与电子商务个性化推荐研究相结合,并在一定的模型假设前提下,提出了基于Agent的餐饮个性化推荐模型架构。(2)针对餐饮个性化推荐理论模型,通过对顾客特征和情境内涵的抽象与界定,区分影响顾客就餐行为的主要影响因素,分别建立了顾客模型及情境模型。(3)在分析本仿真模型的Agent构成及其功能的基础上,重点对顾客及服务Agent的规则库进行设计,通过分析顾客特征及情境因素与顾客行为的关联关系,提炼出基于顾客个性化信息和考虑情境因素的两种规则库,并设计了一个交互管理Agent用于管理各Agent之间具体的信息和行为交互活动。(4)基于REPAST进行仿真模型的设计及实现,重点是代表各Agent的自定义类的设计及实现,其中,顾客和情境模型作为类的属性,规则库作为其方法库。最后,本文将两种规则作为两套运行方案进行实施,从而分别获取顾客特征及情境因素对个性化推荐和顾客行为的微观影响分析,以及最终顾客的宏观涌现行为分析。两种运行方案下模型的有效性评估结果表明：本文提出的基于顾客特征的个性化推荐模型与其它推荐研究的有效性数值范围相当,而考虑情境因素的个性化推荐模型的有效性又有了明显提高。更多还原

【Abstract】 Along with the increasing of information overload problem under Electronic Commerce, personalized recommendation as an effective solution has become one of emerging E-Commerce research areas. With the rapid expansion of mobile network techniques, accurately personalized recommendation considering context under emerging Mobile Commerce is becoming a key technical problem need to be solved urgently.There are several deadly problems in traditional research methods, such as data sparseness, cold start and difficult data collection. Besides, the effectiveness of recommendation strategy needs to be evaluated, and the important impact of context factors on consumer behavior under mobile commerce circumstance needs to be considered. All these problems can be solved by adopting Agent-based Modeling and Simulation (ABMS) method to study personalized recommendation. Moreover, it possesses the ability of seizing emergence of the whole system. Therefore, this paper makes some research on personalized recommendation using ABMS method. Taking catering recommendation system as an example, customer behavior and the effectiveness of personalized recommendation strategy are analyzed according to the emergence generated by the interaction of Agents.Key works of this paper can be summarized in the following parts:(1) By combining ABMD method and personalized recommendation under electronic commerce, the framework of Agent-based catering recommendation model is proposed under some assumed conditions.(2) Aim at the theoretical model of catering personalized recommendation, main influencing factors which affect customer dining behavior are found, then a customer model and context model is established respectively.(3) Based on the analysis of Agent components and their function of this simulation model, focus on the design of rules of customer and waiter Agents, rules based on customer individuation and context factors are proposed respectively through analyzing the relationship between customer characters and context factors with customer behavior. Moreover, an interaction management Agent is designed to supervising the interaction of information and behavior among Agents. (4) The design and implement of this simulation model under REPAST, especially the user-defined classes representing various Agents, where customer and context models as the attributes of classes, rules as their methods.In the end, two operation plans are run based on two kinds of rules to make microcosmic impact analysis of customer characters and context factors upon personalized recommendation and customer behavior, as well as the macroscopic emergence analysis of customer. The effectiveness evaluation results indicate that the customer character-based recommendation model has similar effective value range with other recommendation research, and the effectiveness is improved dramatically in the other operation plan considering context factors.更多还原