Research Of Collaborative Filtering Recommendation Algorithm Based On Fuzzy Theory

With the advent of the era of big data,the increasingly prominent problem of "information overload" has led to a linear increase in the time it takes for people to find content they are interested in on the Internet.The birth of recommender systems has brought a new dawn to their urgent personalized needs,and users have gradually shifted from actively searching for information to passively accepting content of interest.As a classic recommendation algorithm,Collaborative Filtering(CF)has brought a good recommendation effect and also brought huge benefits to the provider.However,with the rapid growth of data in the network,some problems such as data sparsity,poor scalability,cold start and so on gradually appear.This paper makes the following researches on the problems existing in the CF algorithm:(1)Fuzzy set theory is introduced to fuzzy processing the score data,and membership function is used to show the user’s liking degree.Then the user-item fuzzy preference matrix and item-item type matrix are established and combined to obtain the user-item type fuzzy preference matrix.Then the fuzzy c-means(FCM)algorithm is used to cluster the data.Since the number of item types is far less than the number of items,the sparsity of data is alleviated and the space complexity is reduced.At the same time,the FCM algorithm is sensitive to the initial point,and the genetic algorithm is used to optimize the clustering center,which tends to the global optimum.In the similarity calculation,it only needs to calculate with users in the same cluster,which improves the scalability of the algorithm.(2)Similarity calculation is the core of CF algorithm.The quality of nearest neighbor selection directly affects the final recommendation quality.The traditional similarity calculation method only considers the rating data.This paper designs a combination similarity calculation method that integrates user attribute features and item popularity.Firstly,the similarity of user attribute features is calculated,including gender and age attributes.At the same time,the popularity of the item as a penalty factor was incorporated into Pearson’s correlation coefficient.They are dynamically fused and weighted by the threshold of the total number of user ratings.Finally,the score is predicted by the score prediction formula to achieve top-N recommendation.A comparative experiment is conducted on the publicly available Movie Lens dataset.The experimental results show that the designed algorithm has smaller error in the rating prediction task and higher accuracy and recall rate in top-N recommendation than the traditional algorithm.At the same time,a personalized movie recommendation system is designed,and the paper’s algorithm is applied to realize personalized movie recommendation,which verifies the effectiveness and feasibility of the algorithm.Figure [25] table [12] reference [64]...