Dynamic Analyzing And Anomaly Detecting Of Organization Based On Evolution Of Communities

Organization is the collection of members with their attributes and relations for specified purpose.Visible and invisible organizations affect all respect of our lives in the real-world.The anomaly of organization may indicate great actions,changing in evolving or other interesting knowledge.In mining such knowledge,we can get a better view of the development of the organizations focused on and even predict the development direction.Given the sequence of specific organizational networks,this article describe the dynamic of the organization and make three achievements include partition of organizational communities,analyzing of communities evolving and detecting of anomaly sub-sequence.The main innovating points in this article are as follows:(1)Evaluation indicator of number of communities based on F-check.To decide the number of communities is a main problem in community partition,and inspired by the variance analysis in statistics,the F value for community partition is proposed to evaluating differences among communities and serves as evaluation indicator of number of communities.We have verified effectiveness of the F value through the experiments on standard networks in community partition and two real world networks.(2)Indicator for communities evolving based communities' role entropy and a realization of network similarity.The conception of communities' role entropy is proposed using nodes' local clustering coefficient based the thought of information entropy and communities' role entropy is able to reflect the heteroge neity in distribution of roles in communities.Similarity of networks is a common indicator to measure the changes in organizational structure and a cosine similarity measurement is proposed in the article.The effectiveness of both the indicator and the measurement is verified in the experiment of organizational dynamic description.(3)Method for anomaly sub-sequence checking based on neighborhood consistency.The conceptions of sub-sequence and neighborhood sub-sequence are defined and the process of restructure of neighborhood sub-sequence is realized by multi linear regression model.The consistency defined by regression coefficient and restructure bias is proposed to measure consistency between sub-sequence and its neighborhood.There are two methods to calculate the consistency factor including least square estimation and bias optimization.We find consistency factor performs well in find anomaly sub-sequence and provide how to choose parameters.