Research On Hybrid Recommendation Algorithm That Integrates User Features And Item Features

With the rapid development of information technology,the amount of information on the network is growing exponentially,and the era of information overload has arrived.In the face of the problems caused by this,personalized recommendation technology has attracted more and more attention.However,there are still many problems with this technology,such as matrix sparsity,inaccurate similarity calculation results,and cold start problems.The thesis addresses these issues with the following research:First,the thesis proposes a collaborative filtering algorithm that integrates user and item features(UICF).This algorithm improves the following issues:(1)This algorithm optimizes the sparsity problem of the matrix and proposes a matrix filling method that combines item feature preferences.This method fills the matrix by combining the features of the item and the user’s previous ratings.Different filling strategies are adopted based on the degree of difference between item features,and compared with commonly used matrix filling methods,this method has stronger interpretability and effectiveness.(2)This algorithm optimizes the problem of inaccurate similarity calculation results.The algorithm transforms the similarity calculation method based on the classical Pearson correlation coefficient and additionally considers factors such as user activity and item popularity that interfere with similarity calculation to obtain more accurate similarity calculation results.(3)This algorithm optimizes the problem of incomplete consideration of similarity calculation.The algorithm constructs an item feature preference matrix based on the rating matrix and the item features to analyze the similarity of users on the item feature preference level.To mitigate the interference caused by feature biases such as "premature aging" and construct users’ real feature information,the algorithm constructs a user feature preference matrix based on the rating matrix and user features to analyze the similarity of users on the user feature preference level.Both are weighted and integrated into the improved similarity calculation method based on ratings to further improve the accuracy of similarity calculation.The thesis also proposes a recommendation algorithm based on user feature preferences(UFP).This algorithm provides personalized recommendation capabilities for new users by combining the user’s initial features with the user feature preference matrix,effectively alleviating the cold start problem.Finally,the thesis combines the UICF algorithm and the UFP algorithm by switching between hybrid strategies,i.e.,the hybrid recommendation algorithm that integrates user features and item features(UFIF).The thesis conducted five experiments on the MovieLens dataset.By observing and comparing the experimental results,the thesis proves the effectiveness of the proposed algorithm optimizations and the effectiveness of the UFIF algorithm.The thesis also designs and implements a movie recommendation system to verify the practicality of the UFIF algorithm.