Collaborative Filtering Algorithm Based On Social Relationship And Attention Mechanism

With the widespread use of the smart mobile terminal devices and the continuous development of network communication technologies,people can obtain massive amounts of data through the Internet.In the face of the huge amount of data available,the problem that people need to solve is how to efficiently and accurately query the required information.Search engines become one of the solutions by filtering pages that match explicit queries,but it is difficult to give valid keywords when querying.The recommendation system solves these problems to some extent and helps users get useful resources quickly and efficiently in the face of big data.The collaborative filtering algorithm is a widely used recommendation algorithm.It calculates the similarity between users(items)through the user-item rating matrix,generates the nearest neighbor set of the target user(item),and performs score prediction based on the nearest neighbor set.However,collaborative filtering relies heavily on rating information,and there is a problem of sparse data and "cold start".This paper makes improvements to the above problems,introduces social relationship data into collaborative filtering for matrix complementation,and uses clustering algorithm to generate item clustering.In the similarity calculation,the intrinsic attribute characteristics of the item are considered,and the idea of attention mechanism is combined.The main work of this paper is as follows:Firstly,the users' friend relationship data is abstracted into the influence coefficient,and applied to the matrix completion process of the collaborative filtering algorithm,which fully considers the users' social relationship and interest area,and improves the accuracy of the complement score.The collaborative filtering algorithm is used according to the completion matrix,so that the recommendation error of the algorithm is reduced,the recommendation accuracy of the algorithm is improved,and the recommendation result is more suitable for the users' interest field.And K-means algorithm is used to generate item clusters in all item scopes,which reduces the number of item lookups when generating the neighborhood of the items,thereby reduces the number of times of calculating the similarity and improves the efficiency of the algorithm.At the same time,the abstract item clustering has lower sensitivity to the data sparsity of the rating matrix,which ensures the final recommendation accuracy of the algorithm.In this paper,the traditional item-based collaborative filtering,the item collaborative filtering based on mean complement and the collaborative filtering algorithm based on social relationship and clustering are compared by experiments.The advantages of collaborative filtering algorithm based on social relationship and clustering in solving data sparsity problem are verified.It has low recommendation error and high recommendation accuracy,which proves the correctness of the improved thought.Secondly,the attribute feature label of the item is introduced,and the similarity calculation method is improved.The attribute characteristics inside the item are fully considered,so that the similarity calculation between items is more accurate,which makes the recommendation accuracy of the algorithm higher.And combined with the idea of attention mechanism,it gives more attention to items with high similarity to target items,reduces the influence of noise on score prediction,improves the recommendation performance of the algorithm and makes the final recommendation result more in line with the users' actual preference.In this paper,the traditional itembased collaborative filtering,collaborative filtering algorithm based on social relationship and clustering,collaborative filtering algorithm based on social relationship and attention mechanism are compared experimentally.The experimental results show that collaborative filtering algorithm based on social relationship and attention mechanism performs better.