Research On Recommendation Algorithm Based On Grey Relational Clustering

The improvement of information technology and social economy has greatly increased the scale of Internet users,while the presentation of information and content on the Internet has become more chaotic.How to make Internet users quickly and effectively find valuable content in the excessive information is an urgent problem to be solved.In this case,the recommendation system came into being.In the years of development,the recommendation algorithm based on the nearest neighbor has been widely used in various recommendation systems because of its simple and efficient principle.However,the recommendation algorithm based on the nearest neighbor also has some shortcomings,such as data sparsity problem,user cold start problem,user interest drift problem,etc.At the same time,in the similarity calculation,there is also a discernability problem.In view of these problems,this paper finds that the grey system theory has a good advantage in dealing with systems with unclear information by studying and analyzing relevant literature.According to the relevant steps of the grey system theory,the complexity of the system can be reduced by first carrying out grey clustering on the projects or users to reduce the project space,and then carrying out grey correlation analysis on the projects or users in the same cluster to reduce the calculation amount of the system.This paper introduces the theory and proposes an improved hybrid recommendation scheme.The main research work has the following three points:(1)In the recommendation algorithm based on neighbor,similarity calculation method is the important factors influencing the recommended,in this paper,by analyzing the similarity of the traditional calculation method,found that distinguish the sexual problems in calculation,and the grey correlation similarity is made up of some discrete points averaged,instead of the traditional similarity using grey correlation similarity can effectively alleviate this problem.At the same time,because the user's interest will change with time,the time attenuation function is added into the similarity calculation formula to improve the influence of time effect on recommendation.This paper proposes a hybrid recommendation algorithm based on grey correlation analysis by combining two methods of similarity calculation by adjusting parameters,and gives the design architecture and steps of the algorithm.By selecting different parameters for the experiment,when the resolution coefficient is equal to 0.3,the adjustment parameter is equal to 0.6,and the number of neighbors is equal to 40,the recommended accuracy of the proposed algorithm is 10%,6%,and 3% higher than the traditional collaborative filtering,the collaborative filtering based on time effect,and the collaborative filtering based on gray correlation,respectively.(2)Aiming at the problem of data sparseness,this paper uses gray association clustering to merge similar factors,simplify the complexity of the system,and has the advantage of no specific requirements on the sample size and sample regularity.Class hybrid recommendation algorithm.Before using the hybrid recommendation algorithm based on grey association analysis,cluster the items first,reduce the project space,then calculate the similarity within the same cluster,find the nearest neighbors,and finally combine the grey association clustering algorithm with the hybrid recommendation based on grey association analysis.Combining algorithms,the combined algorithm not only has the ability to process sparse data,but also has an improvement effect on the problem raised in(1).At the same time,clustering can be calculated offline,which can significantly improve the efficiency of recommendation.After experimental verification,when the number of clustered neighbors is equal to 50,the proposed algorithm in this paper has the highest accuracy,which is about 3% higher than other algorithms.When the number of clustered neighbors exceeds 50,the algorithm tends to be stable,reflecting the gray-associated clustering pair.Advantages of processing sparse data.(3)Use the MovieLens data set for simulation experiments.First analyze the parameters of the algorithm to find the optimal value,and then perform the experiment again under the conditions of the optimal parameters.Finally,the algorithm of this paper is compared with other algorithms.When the resolution coefficient is 0.3,the adjustment parameter is 0.6,the clustering threshold is 0.7,and the number of neighbors is 50,the proposed algorithm has a slight improvement over other algorithms.The rationality and feasibility of the algorithm.