Research On A Lightweight Collaborative Filtering Recommendation System

Personalized recommendation as an important means of information identification and filtering,brought to the attention of the scholars and the research institutions at home and abroad for a long time.On March 5,2015,China's NPC made the "Internet+" plan,the plan emphasizes traditional enterprises deeply integrated with the Internet,to transform the traditional enterprise and realize the transformation and upgrading of traditional enterprises."Internet+" plan will promote the traditional industry into the electronic commerce field,forming various subdivision of the professional electrical business ecology.The professional electric business platform's process of growth to mature will inevitably lead to items increasing.Consumers will be swimming in the ocean of excess product information.And how to improve the degree of personalization of mining user interest in huge amounts of information has become a breakthrough of promoting customer satisfaction and electricity service level.To sum up,the importance of the personalized recommendation service is increasingly prominent.At present,in the study of personalized recommendation system,collaborative filtering recommendation system has been widely concerned by the researchers which relying on its novelty and accuracy.However,with the advent of the era of big data,and the diversification of users' interests,collaborative filtering recommendation also gradually showed "super high dimension","cold start" problems which seriously affected the recommendation system performance.Based on the problems in collaborative filtering recommendation,this dissertation conducted a lightweight collaborative filtering recommendation system research.This research firstly analyzed the current domestic and foreign research status of collaborative filtering recommendation,and then analyzed the core theory and technology of collaborative filtering recommendation.Later in this article,we established effective forecast user interest's key implicit indicators which based on the implicit behavior,the key implicit indicators can improve the lightweight of the user interest score and avoid "cold start" problem.Finally,based on the key implicit indicators we build the weight in order to form the user implicit rating,and according to the user implicit rating established the preference attributes.We used the preference attributes to calculate the project utility level in order to produce lightweight recommended,avoid excessive project dimension of "super high dimensional" phenomenon which led to the heavy computational burden of the system.Specific studies of this dissertation is as follows: the first chapter is introduction,this chapter mainly elaborated the research background and research significance of this dissertation,and combed the research of the domestic and foreign scholars in implicit indicator and the collaborative filtering recommendation system in order to grasp the overall structure and the direction of the study of recommender systems.Finally,we analyzed the research content and the possible innovations of this article.The second chapter is summary of collaborative filtering recommendation.This chapter aims to analyze the types and features of the user browsing behavior data and describe the core theory and technology of the collaborative filtering recommendation,in order to provide theoretical support and technical support for later research.The third chapter aims to establish user-oriented key implicit indicator.This chapter firstly analyzed the implicit behavior to clarify the category of implicit indicator,and analyzed the collecting and preprocessing methods of the implicit data.Finally this chapter combined with statistical to analysis the validity of implicit indicator in predicting user interests,so as to establish the key implicit indicator in the user browsing behavior.The fourth chapter aims to build lightweight collaborative filtering recommendation system of LW-CF.This chapter firstly build algorithm to establish the key implicit indicators' weights in the user interest,and uses the linear regression model to form the comprehensive implicit ratings.Then based on the comprehensive implicit rating,we designed algorithm to establish the user's preference attributes and their utility level.And attribute utility model was constructed based on the utility level to improve the collaborative filtering recommendation system and form the LW-CF.Finally we used the comparative analysis method to analyze the recommendation accuracy of LW-CF.The fifth chapter is conclusion and prospect.This chapter summarizes the final results of this paper studies and analyses the research direction in the future.Through the research of the lightweight collaborative filtering recommendation system LW-CF,this dissertation introduced the classification of implicit indicator,and based on the statistical analysis method to establish the key implicit indicators which can effective predicting user interest.Finally we used the contrast experiment to verify the accuracy of the LW-CF.The experimental results show that LW-CF has the strong recommendation accuracy.