Research And Application Of Signed Network Community Detection Algorithm

In real complex systems,there are not only positive interactions between members such as support,trust,and facilitation,but also negative interactions such as opposition,doubt and inhibition.Signed networks intuitively and flexibly describe this type of information by introducing positive and negative connections between nodes.Because they contain more abundant complex system information,they are conducive to comprehensively and deeply mining the relationships between individuals.Signed networks have gradually become one of the research hotspots of scholars in various fields at home and abroad.As the basis of signed network analysis,community structure has theoretical significance and practical value in revealing the organizational structure,functional structure,and individual behavior of complex systems.Relevant research results have been widely applied in many fields such as recommendation systems,public safety,biomedicine,and so on.Although there have been frequent studies on community detection algorithms in signed networks in recent years,and certain progress and achievements have been made Most mainstream algorithms have the disadvantages of overly complex mathematical models,high time complexity,and difficulty in applying to large-scale networks,while many mature traditional community detection algorithms cannot be directly applied to signed networks due to the impact of negative attribute edges.Accurate and rapid detection of community structures in signed networks remains a challenging task.To address the above issues,this paper proposes two signed network community detection algorithms and applies them to the field of public opinion analysis.The main research contents of this paper are as follows:(1)From the perspective of simplifying the complexity of mathematical models,this paper designs a normalized Laplacian based Matrix and convex non negative matrix decomposition algorithm(NLRCovex NMF)to detect community structures in signed networks.The algorithm applies convex nonnegative matrix decomposition to signed networks,decomposing the adjacency matrix into community membership matrix and node weight matrix.At the same time,the algorithm introduces normalized Laplacian graph regularization terms to reduce noise generated during the decomposition process.Comparative experiments on artificial networks and real networks have verified that the algorithm has relatively effective community division performance in both small-scale unsigned networks and signed networks.(2)To address the problem that the NLRCovex-NMF algorithm has high time complexity and it is difficult to apply it to large-scale symbolic networks,This paper proposes a graph convolutional aggregation network based on structural equilibrium theory(CNSBGCN).This algorithm combines sociological balance theory with graph convolution neural networks to extract hidden information about graph structures and node attributes,forming an effective node embedding representation.The algorithm designs a loss function based on maximizing the modular degree of the signed network,and completes the community detection task through end-to-end learning.Comparative experiments on multiple network datasets with different characteristics have verified that the algorithm can detect a more reasonable community structure on larger signed networks.(3)Apply the theoretical results to the project of "Risk Public Opinion Discovery and Disposal",combined with the relevant content data,forwarding and comment information of Sina Weibo in a public opinion event of the new crown epidemic that was publicly detected.Calculate sentiment score,build signed network in time period,The algorithm proposed in this paper is used to explore the characteristics of core group structure changes in the evolution of public opinion.While verifying the effectiveness of the proposed algorithm,Provide data analysis auxiliary support for in-depth public opinion discovery,public opinion evolution trend prediction,public opinion guidance target group identification,etc.