The Research On The Recommendation Algorithm For The Online Social Network Based On Community Detection

With the increasing development of the Internet,especially the growing popularity of mobile Internet,the information in the network presents explosive growth so that information overload is increasingly serious.The recommendation system is one of the important solutions to the overload problem which taps users' interest by an in-depth analysis on their historical behaviors and transforms autonomous search by the users themselves into personalized information service,thereby effectively improving their efficiency of information acquisition.Communitization is one of the important characteristics of the online society.The internal vertexes of the communities closely link with each other.Meanwhile,there are only a few connections between any two communities.On the one hand,a network hierarchy can be effectively explored in virtue of this feature.Research have found that the combination of community detection and the recommendation algorithm leads to a well positive evaluation for its improving the recommendation performance of the recommendation algorithm.This paper takes the CosRA algorithm as an example and optimizes the CosRA algorithm by combining it with the community detection to the improve the upper bound of the CosRA algorithm performance.The main tasks are as follows:1.Use the Jaccard similarity theory to build connections between products and construct user behaviors into product connection diagrams suitable for the community detection.2.According to the vertex density,divide the products into corresponding communities,and further optimize the result of the division by local modularity,thereby obtaining a better community detection result suitable for the use of the recommendation algorithm and gaining user-division sets by inverse derivation.3.Combine the community detection with the recommendation algorithm and give suggestions for the final recommendation result screening.Validate the accuracy and novelty of the algorithm by public data sets,analyze the effectiveness of the algorithm under the optimal and average conditions,and analyze the screening of the recommendation result by experiments.The algorithm introduced in the article has an excellent effect with the priority given to accuracy,and the average effect of large data sets is better than that of the small ones.In the large data sets,the effect of the algorithm in the data sets with higher user-product ratio is better than that in the data sets with lower user-product ratio.This phenomenon also conforms to the general usage scenarios of the recommended algorithm.