Research And Application Of Community Detection Algorithm Based On Probabilistic Graphical Model

With the development of computer science and technology,how to find the logical relationship between data from complex data and networks,and abstract the data value of the network is a very important topic.In recent years,a large number of community detection algorithms have emerged in the literature to divide the network into different community structures.The community detection method based on Markov random field has great potential in the detection of community structure in complex networks,but there are still accuracy and The efficiency is not high,the model selection is over-fitting or under-fitting,and the algorithm complexity is high.Therefore,this paper combines the probability graph model and the belief propagation algorithm,and proposes two improved community detection algorithms based on the probability graph model,and applies the algorithm to practical projects.The main research work is as follows:(1)Propose a local optimization algorithm LOSN-BP based on seed nodes and belief propagation,use the improved heuristic algorithm KSIE in this paper to select seed nodes,and reduce the problem of low accuracy caused by single index selection of seed nodes;community In the expansion stage,the modularity function is mapped to the energy function,and the message propagation mechanism in the BP algorithm is improved.With the seed node as the root node,a probability graph model is constructed along two layers of neighbor nodes to propagate,and the node with the highest edge probability is selected to be incorporated into the community;In order to ensure the convergence of the algorithm,this paper uses the Laplacian matrix in computational geometry to adjust the probability graphical model to ensure that the algorithm can converge to the correct result within a limited time;in order to improve the accuracy of the algorithm,after the community division In the processing stage,this paper re-divides the community based on the strategy of maximizing modularity and community similarity,and further optimizes the quality of community division.Comparative experiments on different types of network datasets verify the effectiveness of the LOSN-BP algorithm.(2)A global optimization algorithm GOMK-BP based on Markov and belief propagation is proposed.This algorithm maps the similarity and importance indicators of nodes to energy functions,and at the same time optimizes and improves the performance of the model,combining the form of exponential function and weighted sum Make the algorithm achieve a better model fitting effect;the probability graph model used by the algorithm is a fully connected Markov random field,because even if there are no edges between nodes in the actual network,there will be mutual influence between them;In order to improve the efficiency and accuracy of the algorithm,this paper improves the traditional message propagation algorithm and uses compressed sensing to accelerate the convergence of the algorithm;because the fully connected Markov random field may be affected by high computational complexity,this paper adjusts The model structure is improved,and sparse connections are introduced to balance computational efficiency and model performance.Verification on multiple groups of artificial networks and real networks shows that the GOMK-BP algorithm can indeed detect high-quality community structures.(3)Apply the algorithm proposed in this paper to the project ”Literature Network and Discourse Analysis of Academic Achievements in English and Central Asian Studies”,and build a network according to the citation relationship between literature or authors in a certain field.Through the algorithm proposed in this paper,we can further explore the research hotspots,trends and core areas in this field,and identify the leaders of academic discourse in this field,so as to effectively serve the analysis of discourse patterns and assist academic research.