Research On Agent-based Catering Personalized Recommendation Modeling And Simulation

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.