Personalized Recommendation Method Based On Fuzzy Theory And Collaborative Filtering

Along with the development and perfection of network technology, many personalized recommendation systems emerge for better user experience. One of the common recommendation technologies, Collaborative Filtering Recommendation algorithm is hindered by these bottlenecks, such as sparsity, cold start and scalability, in recommendation.Traditional personalized recommendation algorithms tend to ignore the fact that similar taste often only exists in one respect between two individuals in real life. In addition,sparsity problem, inherent in Collaborative Filtering Recommendation system, greatly influenced the recommendation accuracy, through making locating the set of nearest neighborhood user difficult.In this thesis, improving certain defects of Collaborative Filtering Recommendation system, which is widely used in audio visual field, aims to dig deeper into the potential interest of user. In consideration of rating preference and interest migration, the recommendation result of more accurate and diverse can be obtained.On the basis of above consideration, a personalized recommendation method based on Fuzzy Theory and Collaborative Filtering is presented. Firstly, data mining techniques based on fuzzy theory, developed rapidly in recent years, are introduced to provide technical support for the improving of the recommendation system. Secondly, common personalized recommendation systems are described. On the basis of the current movie recommendation system, the advantages and disadvantages of Collaborative Filtering Recommendation algorithm should be understood and mastered. What's more, new mathematical models are proposed to improve the traditional collaborative filtering algorithm. At last, a complete system architecture of the Personalized Recommendation Method Based on Fuzzy Theory and Collaborative Filtering is provided on the basis of the original work.The main work of this thesis is divided into the following areas: Above all, on the basis of Concept Hierarchy, adopt the concepts of “membership” and “trustworthy”, the Fuzzy Support Vector Machine is used to fuzzy-clustering the items, and the Fuzzy C-means Classification is used to calculated the trustworthy of user for each type of items. Through the simple dimension reduction and reduced data set by using Fuzzy Clustering analysis and Fuzzy Classification, the improved personalized recommendation method can both to improve the sparsity problem and to improve the scalability of algorithm by reducing the search scope. Followed, the concepts of “rating preference” and “interest migration” are adopted into new user similarity measure formula and rating prediction formula. The accuracy of the improved Collaborative Filtering has a certain degree of improvement. At last, Improved Collaborative Filtering Recommendation algorithm is used to calculate the recommendation items, respectively corresponded to every item class. Taking relevant information onto account, the new defined weighting model is used to sort out the list of all recommendation items.Experimental results show that the improved personalized recommendation method may expand the diversity of the recommendation result and further explore the potential interest of user. Meanwhile, the new personalized recommendation method can obtain a better result with more accuracy, more timeliness and more interpretability.Without considering the case that part of users do not rate for items, the downside is assuming all users will directly rate for their browsed projects. Estimating the ungraded items will be attempted firstly in the following study. Mathematical models do not have enough support, it needs to be further optimized and improved. In addition, large number of off-line computing requires regularly updated database, which will lead to real-time problem, how to improve this question is another direction for future research.