A Smart B2C E-commerce System Based On ACP Approach

E-commerce is becoming a more and more important channel of retailing.Currently, it is an era of information overload, and users of e-commerce sites urgentlyneed anefficient and accurate way to find what theywant.Such demand makesrecommendation system arise.In order to improve the turnover, e-commerce sites aretrying to develop better recommendation system to stimulate users to make moreconsumption, and to improve the loyalty towards the web site.Recommendation methods are generally divided into3categories,content-based,collaborative filtering and hybrid recommendation.Traditional recommendationmethods based on historical purchase data and rating data has good prediction accuracy,but with the expansion of data size and rating matrix becoming sparse, the accuracy andresponse time become far from satisfactory.In the literature review section, we analyzethe advantages, disadvantages and applicable conditions of various recommendationmethods.Considering the inherent limitation of traditional methods, this thesisproposesasmart recommendation method based on ACP approach.Applying multi-agentsimulation, we build an artificial e-commerce system. We change configuration ofagents in the artificial system through observations in the real system. We obtain agents’purchasing results by multi-attribute decision making theory and provide them asrecommendation results. The third to the fifth chapter of this thesisrespectivelyintroduces the modeling process, algorithm design and experimental platform design.The method we purpose more accurately simulates users’ decision-making behavior,keeps track of user preferences changes and hasimmediate response to environmentalchanges. Data test and analysis of experimental results show that this methodoutperforms traditional methods in terms of user satisfaction and response speed.In the end, we summarize the unique advantages of the method, and put forwardthe future research directions.