Improvement And Application Of LFM Community Detection Algorithm

As the development of information technology,the big data era is coming.A wealth of valuable information,hiding behind massive amounts of data,needs to be extracted.Massive amounts of data leads to complex systems.Complex network theory is an effective way to study on the complex system by constructing network model and analyzing the structure of the network.As an important feature of complex network,the community structure is useful for deep understanding of complex network.Community detection is significant for the mining of hidden information in network.Focusing on the summarization and heritage of TCM experience,the project “Study on the regulations of the modern prominent veteran TCM physicians' experience in syndromes-prescriptions-drugs for COPD based on complex network method”,supported by National Natural Science Foundation,analyses and studies on TCM data,using complex network theory.It's a project to solve the problem of TCM experience summarization.Based on the relationship among system factors rather than system factor itself,TCM theory is full of complexity,relating to pathogenesis,symptom and the rule of TCM formulae.So,taking complex system theory as the object of study,complex network theory has practical significance to the deep exploration of TCM experience.Taking the COPD data as an example,this paper applied community detection algorithm into the exploration of TCM formulae rule.The TCM network is a dense and weighted network with overlapping community.According to these characteristics,this paper selected LFM algorithm,an overlapping community detection algorithm based on local fitness maximization,as the object of study,which has lower time complexity.In order to solve problems existed in LFM,improved algorithms were presented and validated.The main works and innovations are as following:1.After summarization and comparison of existing community detection algorithm,problems existed in LFM algorithm are analyzed,including algorithm instability,the problem of abnormal community and decline of algorithm precision in network with fuzzing community structure.2.LFMd(LFM with Dynamic Parameter)algorithm is proposed.In this algorithm,the Jaccard similarity coefficient is used to make the network structure more clear;the network clustering coefficient is used to improve the community size parameter and make expansion of community more reasonable.LFMd algorithm raised the recognition rate of overlapping nodes in star network and improved the accuracy of traditional LFM algorithm.3.LFMc(LFM based on Clique)algorithm is proposed.In this algorithm,random walk theory is used to measure the similarity of vertex,which made the information of network weights fully utilized;maximum clique theory is used to expand community.LFMc algorithm not only improved the accuracy of traditional LFM algorithm,but also keeps LFM suitable for weighted network and avoids the problem of abnormal community.4.Taking Chinese medicine formulae of Deng Qiyuan as an example,the above algorithms were combined and used to detect communities in COPD network.Some tentative rules of TCM were found.After confirmed by TCM experts,the results are consistent with TCM theory.