Research On Recommendation Algorithm Based On Collaborative Filtering Technology

With the rapid development of the information age,the problem of information overload has become increasingly serious and has affected people's daily lives.Therefore,the researchers proposed a recommendation technology,and the recommendation system provided users with personalized recommendations.How to provide users with more accurate and more personalized recommendations is the current research direction of recommendation algorithms.Collaborative filtering recommendation algorithm is the most widely used recommendation algorithm for Internet companies.Although collaborative filtering recommendation algorithm is widely used,as the amount of data of users and commodities continues to increase,the problems associated with collaborative filtering recommendation algorithms have become more and more serious.The recommended quality of collaborative filtering recommendation algorithm is gradually declining.If this problem can be solved effectively that improve the user experience,but also bring more profits to the company.Based on the above background,the research of this thesis mainly focuses on the problem of data sparsity existing in traditional algorithms.User's rating information and user attribute information are used to establish user's mixed similarity.And for the traditional algorithm does not consider the impact of time on the recommendation results,proposing a time attenuation factor model,combined with the user rating to get the actual user rating,improve the accuracy of the recommendation results.Second,the traditional collaborative filtering recommendation algorithm must traverse users in the entire recommendation system when calculating the user's neighborhood users.However,it is very difficult to traverse all users in the system in practical applications,and the feasibility is extremely low;and the selection of the number of users in the neighborhood is also a problem that needs to be solved.By using the clustering technology,the search range of the neighborhood user can be narrowed down to a user with higher similarity with the target user,so that the efficiency of the search for the neighborhood user can be improved.Aiming at the problem of inaccurate clustering results caused by the traditional initialization of the cluster center by the traditional K-means algorithm,a K-means clustering algorithm(OICK-means for short)that optimizes the initial clustering center is proposed.The hybrid similarity is used instead.In Euclidean distance,users with higher similarity are clustered in the same cluster,and then the favorite items of the target user are predicted based on the history of the neighboring users and Top-N recommendations are made.Experiments on the MovieLens-100 K dataset show that the OICK-means recommendation algorithm based on mixed similarity is superior to the traditional collaborative filtering recommendation algorithm in all evaluation indicators and has a better recommendation effect.