Research On Hybrid Recommendation Algorithms Based On Clustering And Trust Relations

With the rapid development of the Internet,network information is also increasing,and information overload is becoming more and more serious.Faced with the huge amount of information,how to quickly discover rules and provide valuable information for users has become an urgent problem to be solved.Recommendation system emerges as the times require.As the core of recommendation system,recommendation algorithm is an important factor affecting users,satisfaction with recommendation system.At present,although the commonly used collaborative filtering recommendation algorithm has its own advantages,there are still some problems,such as the low efficiency and accuracy of the recommendation algorithm due to the high sparseness of the data itself.Therefore,this paper improves the algorithm from multiple perspectives.The specific improvement strategies are summarized as follows:Firstly,in order to improve the computational efficiency of the algorithm,K-means clustering algorithm is introduced in this paper.Before searching the nearest neighbor set,users are clustered in advance so that similar users can be found in the same cluster.When searching the nearest neighbor of the target user,they can search directly in the cluster where the target user is located,which saves the searching time effectively.At the same time,considering that in the traditional K-means clustering algorithm,it is not scientific to select the initial clustering center randomly,so this paper improves the selection method of the initial clustering center.After selecting an initial centroid randomly,the farther away from the initial centroid,the more probability it will be selected as the next centroid.The residual centroid is selected,which improves the traditional K-means algorithm.The instability improves the overall effect of the recommendation algorithm.Secondly,in order to solve the problem of low accuracy of recommendation algorithm,we optimize it from the following aspects:1)When traditional recommendation algorithm calculates similarity,because of sparse data,the calculation of similarity is inaccurate,which leads to the problem of low accuracy of recommendation.Based on the clustering results,Slope One algorithm is used to fill the score matrix and transform it into several dense small matrices to reduce data sparsity.2)This paper introduces the concept of trust relationship,which is used to calculate the trust relationship between two users who do not overestimate.In this paper,we discuss the direct trust,indirect trust and user's own reputation among users,and use the above three trust degrees to construct user's final trust as part of the subsequent calculation of similarity.3)Considering that users'preferences will migrate over time,this paper introduces the concept of user interest stability,incorporates time-sensitive factors into the calculation process,and uses variance to measure the stability similarity of users' interests;and then fuses the initial similarity between trust and user interest stability,and acts as the distance between users in K-means clustering,and at the same time with the distance through K-means clustering.-The final similarity is obtained by fusing the similarity of score matrix filled with menas clustering and Slope One.According to the final similarity,the nearest neighbor of the target user is searched from the cluster where the target user is located.Finally,the score prediction is made and the recommendation is made.