Research And Application Of Link Prediction Based On Community Relationship Strength And Node Attributes

The main task of link prediction is to predict the possibility of connecting edges between nodes in the real network,which has important theoretical value for analyzing the evolution mechanism of the network and reconstructing the network,and has a wide application value for understanding protein interactions,social platform friend recommendation,and personalized product recommendation of e-commerce websites.The traditional link prediction method predicts the possibility of linking between nodes based on similarity indicators,ignoring the impact of community structure information on the links between nodes,resulting in unsatisfactory prediction effects.Aiming at the above problems,this paper uses the community detection algorithm to identify the community structure of complex networks,integrates the strength of community relationships and node attributes,conducts in-depth research on the prediction method of complex network links,and applies them to personalized music recommendation system.The main work of the thesis is as follows:(1)Aiming at the problem that the local community division algorithm based on label propagation is not accurate and the stability is poor,the label propagation algorithm is combined with the community detection algorithm(CD)to establish a community detection algorithm based on label propagation LPAIS＿CD.The algorithm consists of two phases,partitioning and union.In view of the CD algorithm,which only considers the community information and does not consider the impact of the nature of the node itself on the community,resulting in low accurac y of community division,a calculation method of label influence is proposed in the partitioning stage combined with the importance and similarity of nodes,and the network is divided into multiple communities according to the label influence.In order to improve the accuracy and stability of the detection algorithm,unstable communities that are not sufficiently isolated are merged with other suitable communities in the zone phase.LPAIS＿CD algorithm integrates the node’s own attributes and community information,which can improve the accuracy and stability of community division.Experiments were conducted in 2 artificial networks and 9real networks,and the results showed that LPAIS＿CD community detection algorithm can obtain a better community structure,and the structure information in the community can be extracted more effectively.(2)Aiming at the link prediction algorithm based on community information,which only considers the link influence between nodes in the same community,ignores the influence of community relationship on link prediction,and integrates the relationship strength and node attributes of the community,a link prediction algorithm CDICRS that can adapt to overlapping and non-overlapping communities is proposed.Firstly,the original network is divided into multiple communities using the LPAIS＿CD community detection algorithm,and then a community relationship metric is proposed to measure the relationship strength of the communities in the network,and the community relationship strength index is fused with the improved similarity index to estimate the connection likelihood score of the unconnected node pair.Experiments were conducted in 10 real networks of different sizes and sizes,and the results show that the proposed algorithm has better link prediction effect and lower time complexity.(3)The proposed CDICRS link prediction algorithm is applied to song recommendation,and a personalized song recommendation prototype system is realized.Firstly,the user-song similarity matrix is obtained through the user behavior log,the user-song network structure is constructed,and then the CDICRS algorithm is used to calculate the likelihood score between the user and the user,the song and the song,and the user and the song,and finally the personalized song recommendation is based on the user’s similarity set’s preference for the song and the song similarity set.