Research on Collaborative Filtering Recommender Algorithm Based on Clustering and Preference of Item Categories
|School||Zhejiang University of Technology|
|Course||Management Science and Engineering|
|Keywords||Personalized recommendation Collaborative filtering Scalability Datascarcity Item category preference Clustering|
In order to solve the information overload problem, recommendation algorithmand personalized recommendation system become the focus and difficulty of theresearch and application of various. Against data scarcity, scalability problems, coldstart and synonyms problem of traditional collaborative filtering algorithm, in order toimprove the quality of the target user’s nearest neighbors as the starting point, thepaper proposed an improved CF algorithm in view of the existing data sparsenessproblem and system scalability problem of the traditional CF algorithm based on user,and carried on the empirical analysis to verify the validity and feasibility of theresearch results in this paper.The main research work of this paper is as follows:Chapter I: Introduction section. Introduced the research background, researchstatus and existing problems, the main research content and research ideas, structuralarrangements of this paper.Chapter II: Personalized recommendation system and related technologies aboutrecommendation. Briefly describes the classification and system architecture ofe-commerce recommendation system, mainly introduced content-based recommendationtechnology, collaborative filtering technology and its recommendation principles andprocedures.Chapter III: Users clustering based on the K-means and project categorypreference. In view of the scalability problems of collaborative filteringrecommendation, introduces the application of clustering algorithm in collaborativefiltering recommendation, proposed a clustering algorithm based on K-means andproject type preference. The algorithm is used to cluster users offline, divide users ofsimilar project type preference to user clusters with the same and then thecollaborative filtering recommendation based on clustering can find neighbors fromseveral clusters similar to target user, improves the query efficiency.Chapter IV: Collaborative filtering algorithm based on Clustering and projectpreference category preference. In view of poor recommendation quality of traditional user-based collaborative filtering algorithms in the case of data sparseness datasparseness, this chapter firstly put forward a model of collaborative filteringrecommendation based on the project category preferences. The recommender modelconsiders not only rating information of users but also project category preferenceinformation of users when determining neighbor collection of the target user, whichhelps to find neighbors more accurate and improve the recommendation quality ofsystem. Secondly proposes the CF algorithm based on Clustering and projectpreference category preference by combining the clustering algorithm proposed in thethird chapter with the model and gives its recommendation process.Chapter V: Experimental design and analysis of results. Based on MovieLensdata sets, the proposed algorithm is simulated and tested, compared with thetraditional CF algorithm, the proposed algorithm has good query efficiency of nearestneighbors, and to some extent alleviate the scalability problem of the traditionalcollaborative filtering algorithm, effectively solve recommendation quality problemsof traditional collaborative filtering algorithm under the condition of data sparsenessand high-dimensional data.Chapter VI: Conclusion and Outlook. Make a brief summary of the mainresearch contents and results of the paper, and put forward some prospects of furtherresearch.