Research On Collaborative Filtering Algorithm Based On XML Fuzzy Data

With the continuous expansion of network,people's requirements for personal information are getting higher and higher,in order to improve the quality of information,we need a customized recommendation system which can help users quickly search for the most suitable products by analyzing user's preferences.Collaborative filtering technology is the most successful technology.Although collaborative filtering has proven successful and widely accepted,it also has problems such as lower accuracy,data sparsity,and cold start in new item.This paper analyzes the limitations of the traditional collaborative filtering algorithm,proposes a collaborative filtering recommendation algorithm based on XML fuzzy data.Aiming at the sparsity of scoring matrix in the collaborative filtering recommendation algorithm,this paper takes "User's preference can be decomposed into the sum of the scores of each attribute characteristic in the item" as the starting point.These missing values in the sparse matrix are filled by logistic regression model,and the collaborative similarity between items is calculated based on the filled matrix.The processing of the fuzzy data is combined with the recommendation mechanism.Using possibility theory and similarity theory to process and model the fuzzy attribute characteristics of items.This method extracts the similarity between the fuzzy attribute characteristics of items through the good structure of XML,which improves the recommendation accuracy.For the cold start problem in new item,this paper combines the weighted fuzzy similarity of items with the collaborative similarity of items to obtain a comprehensive similarity relationship.Finally,the recommendation is made through this comprehensive similarity.MovieLens data set is used to evaluate and analyze the improved algorithm.Experiments show that under the same experimental conditions,the improved algorithm calculates the similarity between the fuzzy attribute characteristics of items and fills the sparse scoring matrix to make the calculation of the neighboring items more accurate.Accuracy and mean absolute error are better than the comparison of collaborative filtering algorithm.All in all,the proposed algorithm improves the accuracy of the recommendation system,and effectively solves the problem of data sparsity in recommendation system and cold start in new item.