Research On Context-based Personalized Commodity Recommendation Methods

With the rapid development of electronic commerce and the Internet, and the advent of the era of "big data", there has been an increasing of variety and number of information resources which are presented in front of users. And as the gradual increase of the user behavior data generated and accumulated in daily operation, it compounds the difficulties of screening and getting valuable information resources from large number and wide variety of resources. Under these circumstances, it has posed great challenges to the personalized recommendation service on how to find the differences and relations between these information resources, to analyze the users’requirements effectively and help them to find interested resources from a booming ocean of information. With the widespread use of e-commerce and its further complication, it has become a new trend of personalized recommendation research to study how to accurately recommend information resources according to the user actual requirements under different contexts.The user preferences mining and recommendation technology are the most basic and critical work in the task of personalized recommendation, and their quality is directly related to the result of the personalized recommendation. User preferences are broad, and they may be constantly changing with the change of environments. In addition, as the number of customers and information resources increase, and the characteristics of the context is complicated, the results of different recommending methods need better interpretation. Therefore, this paper mainly focuses on the commodity personalized recommendation in e-commerce, analyses the relationship between user and contexts, and discusses how to effectively connect user preferences and commodities in order to recommend commodities to users, which best meet their preferences. In this dissertation, the main research includes the following aspects.(1) Researching the knowledge representation of personalized recommendation. In the light of the connotation and trait of personalized recommendation in electronic commerce, we use the ontology modeling approach to propose an incorporating context commodity personalized recommendation knowledge model. The model effectively expresses the concept of users, commodities and contexts and the semantic relationships between them, it supplies the knowledge support for the implementation of personalized recommendation on concept semantic level.(2) Analyzing user preferences on the basis of user contexts. The research proposes the user preference Bayesian model to express the relationships between user preferences and user contexts. Based on this model, we use the probability reasoning method of Bayesian network to analyze the user preferences under some specific contexts. In addition, on the basis of the information entropy theory, this study puts forward the concept of context information entropy to measure the user’s choice behavior of commodity resources in various contexts, and then judge the importance of the various kinds of contexts, and further analyze the contextualized user preferences according to the context importance.(3) The designing of commodity personalized hybrid recommendation method based on context. This study adopts the contextual modeling recommendation method, from the perspective of user preferences for commodity attributes, and makes exploratory study of the commodity personalized recommendation method which based on the context. This hybrid method is mainly divided into two stages:collaborative filtering and knowledge filter. First, based on user ratings and user preferences for commodity attributes, search the neighbor users, and blend the similarity matching and the context importance in the recommendation generated process of collaborative filtering, to generate the recommendation results; Then, according to user’s context and the contextualized user preference, we adopt the method of knowledge filter in personalized recommendation knowledge model to reason out the commodity resources conforming to the current context and generate recommendation result, and adopt the reasoning optimization method based on contexts to handle the conflict with the collaborative filtering recommendation result, and then form the final results to the user.(4) Application research. At last, the proposed methods in previous sections are applied to the catering personalized recommendation of mobile commerce. This study focuses on the catering dishes recommendation service of mobile commerce, designing the system framework of the context-based catering service, and then analyzing the realization process of personalized catering dishes recommendation in mobile commerce on the basis of a practical case.