Research On Key Graph-based Community Detection For Public Security Intelligence

In the mining of social relation networks, in order to classify a large number of complex populations, we often need to find different groups from the social networks, and according to the individual characteristics or the characteristics of social accounts paying attention to each other doing the clustering analysis is a commonly used solution. However, under the public security intelligence scene, the character of people's behavior is the information that intelligence personnel are very concerned about, and it is new ideas to achieve effective clustering analysis that how to quantify and cluster the characteristics of human behavior, leading to the discovery of the crowd of similar behavior characteristics and classifying them.In the premise of full understanding public security intelligence service, we put forward a KeyGraph-based community detection algorithm(KCD, KeyGraph-based community detection), in order to find out the potential relationship between groups and provide decision support for the Department of public security intelligence. KCD try to start from the behavior characteristics of human to do community detection by establishing KeyGraph and using graph clustering algorithm. Firstly, KCD quantify the behavior characteristics of multiple dimensions between person and person and merge quantified values of multi-dimensional behavior characteristics to form three tuple "people- people-value" set of co-occurrence; then it read the co-occurrence set, filtering out the noise data, establishing the undirected graph of human behavior characteristics; finally it applied the graph clustering algorithm SCAN on undirected graph to find out a number of different groups. In order to deal with large scale graph, SCAN algorithm is improved to obtain SparkSCAN and SparkSCAN can not only run parallel in spark platform, quickly and efficiently completing graph clustering, but also find hubs and outliers, solving the problem of mining the key person between groups in public security intelligence scene.The experiments of KCD using desensitization behavior data in public security intelligence scene are performance comparison test of data partition and storage model, parameters test of improved clustering algorithm, performance test between SparkSCAN and SCAN etc. The results show that KCD can effectively solve the problem public security intelligence encountered of doing community detection using the behavior data, the use of data partition and storage structure of semi-structured shortening computation time of cooccurrence degree, and in the set of behavior data under public security intelligence scene, algorithm SparkSCAN improved by an average of about 69.9% than algorithm SCAN in execution efficiency, which can effectively deal with problem of graph clustering on largescale undirected graph, and improvement is more and more obvious with the expansion of the size of the data set.