Study Of Hybrid Recommendation Model And Strategy Of The Online Shopping Based On Online Reviews Mining

With the rapid development of the web2.0, network data is geometric growth. Obtain the user demand information rapidly and accurately becomes the urgent needs of the related enterprises and customers. A variety of recommendation systems arise at the historic moment. Traditional recommender systems mostly set products as the center based on user ratings. The performance of the system relies on the quality of user preference model. But the user preference information is very difficult to simple score to comprehensive describe by simple user ratings. Therefore, recommender systems often appear cold start and sparse data problem.This article put forward online reviews as data source. Based on analysis of related theory, technology and method of the traditional recommender systems of online shopping and online reviews:Firstly, Online reviews as the main communication forms of internet word-of-mouth, it always was priori as exogenous variables in the study of existing literature,and looked the dynamic relationship as a static single direction between them. Under the dynamic endogeneity, the control variables that can be measured and dummy variable of difficult to observation and measurement outside of online reviews were introduced. This article researched perception problem about products internet word of mouth of consumers based on online reviews. In the dynamic panel data model, the endogenous was controlled by the control variables and the dummy variable, this article demonstrates:(1) In the static analysis framework, online reviews and iwom perception was influence each other. Dummy variable would impact online reviews and iwom perception at the same time.(2) In the dynamic analysis framework, it was found that there existed dynamic influence between online reviews and iwom perception. But the lag phase was not sure. And feedback effect did not exist between the proportion of negative reviews and iwom perception. This shown that the intertemporal dynamic role was not each other but in a single direction. Through the analysis of the influence factors of iwom perception, this article confirmed the different influence from each attribute of online reviews to consumer perception of iwom and identifies the key factors. It provided the basis for differentiation mining of online reviews information.Secondly, on the basis of the above analysis, this article mainly focused on the research of mining of online reviews, including two parts the data source mining of online reviews and information mining of online reviews. Unlike previous online review analysis, the data directly take from the network shopping platform or professional review sites. This study regarded the Internet as data source of online reviews, and excavated reliable data source. Through resolve the research into three subtasks, starting from operation the improved Page Rank eliminate cheating page of network. Using the improved TC-Page Rank to refine web page of high topic relevance and contains a large number of online reviews data. The research used the improved HITS to determine the authority web page of online review analysis data source at last. To the research on information mining of online reviews, online reviews as an important reference for potential consumers to online shopping, mining the valuable information was the key to effective use of online reviews. In view of the design principles of online shopping platform and the actual demand of consumers, this article fused social tagging to construct domain ontology. Based on domain ontology, feature words of online reviews were mapped as ontology concept. And Jess reasoning engine was used to extract the implicit product attributes from reviews. Then this article constructed the hierarchical product attributes after the hierarchical relationships between concepts were mapped to product attributes. Based on cascading CRFs model and emotional dictionary, the hierarchical analysis of emotional orientation was achieved from the polarity analysis of online reviews to the emotional intensity analysis of sentence level to the appraisement intensity analysis of product attribute level.Finally, for the data sparsity, cold start problem were increasingly prominent and the deficiency of user preferences information collection based on rating information, which leaded to the recommended effect of recommendation algorithm could not be satisfied by users. Based on the above analysis of the influence factors of iwom perception, this article presented a construction method of user preference model and product feature model based on information mining of online reviews. Based on the hierarchical user preference information, this article further built the dynamic user preference model by ontology modeling method. And through correction ontology of user preferences for the increase, cut and adjustment user preferences to keep the ontology of user preference dynamic update.Before constructing hybrid recommendation model of the online shopping, this article researched how to design the recommendation system in order to gain the trust of the user, then to realize the expected effect of recommendation system. Used the related theory and method of management, psychology and informatics, and based on interpersonal trust theory to divide the process of user’s trust in the recommendation system into three stages from initial trust to interactive trust and to recommend trust. This article discussed the key factors which affected trust of each stage and constructs the comprehensive model of multistage user trust. Based on the study, this article analyzes the key influence factors of user credible and adopted recommendation system. On the basis of empirical research, according to ISDT framework which was put forward by Walls, this article elaborates the recommendation system characteristics of the user perception credible and adopt from two aspects of Meta requirements and Meta designs respectively.Based on the above analysis, this article built the hybrid recommendation model of the online shopping based on online reviews mining. The model refined the recommendation to the level of product characteristics, and comprehensive sorting according to consumers’ evaluation of product characteristics. The model used the collaborative filtering algorithm as the framework, combined with content-based recommendation algorithm. It eased data sparsity through multi-attribute rating. And this article solved the user cold start and product cold start problem to a certain extent through the calculation algorithm of user attribute-based similarity and product attribute-based similarity. Finally, this article combined with multiple similarity algorithms to construct hybrid recommendation algorithm based on user preference and product feature.Simulation experiments verifiy the good recommendation accuracy and the ability to solve the cold start problem of the hybrid recommendation model of the online shopping based on online reviews mining through collecting 10000 online reviews information from the mobile channel of Taobao, Amazon China, and Jingdong which are the three large domestic network shopping platform.Based on the above research results, this article discusses the recommended strategy of recommender system of online shopping and the main countermeasures and suggestions of online shopping platform in product marketing management practice. Based on the full research, this article summarizes the main research contents and contributions, and expounds the deficiency as well as on the outlook for further research.