Design And Implementation Of Group Detection System Based On User Profiles

With the rapid development of social networks,the way people live and communicate has changed,and new forms of crime have been provided for illegal groups.Through accurate,effective and timely analysis of the internal relationships between illegitimate groups and gangs,targeted individual and group can be quickly identified.However,social networks are complex,information is scattered and inaccurate,and it is difficult to locate the social entities behind a virtual network account.Therefore,it is of great theoretical significance and application value to design and implement a group detection system based on user profiles by collecting and mining the massive data on social networks.Firstly,by study of the user profiles models,we know that user d ata from different sources on the social platform contains different characteristics and the existing user profiles models predicts different tasks separately,ignoring the relationship between tasks.Therefore,a multi-source and multi-task learning user profiles model is proposed by this paper.By using deep learning to integrate multi-source heterogeneous social information,comprehensively explore the effective features implicit in it.Users are fully described in terms of gender,age,region,education level,and topic.Through the connections between the various attributes,a multi-task learning framework is built to learn multiple tasks simultaneously.Experiments show that,the average accuracy of the model proposed by this paper on the real dataset can reach 83%.Secondly,by study of the group detection algorithm,we know that the existing group detection algorithms divide the group through the social network s structure,ignoring the user's own attributes,or grouping by user attribute similarity,i gnoring the social structure.And the user's attributes are numerous and cannot be enumerated one by one.Therefore,a group detection algorithm based on user profiles is proposed by this paper.This paper first calculates the user's comprehensive similari ty based on the fusion vector in the user profiles shared layer.Then,the k-nodes closest to each node are found based on the comprehensive similarity,and the weighted network diagram is reconstructed.At last,the W-Louvain algorithm proposed by this paper is used for group division.Experiments show that,compared with other group detection algorithms,the algorithm proposed by this paper performs best on all evaluation indicators.And it can divide users who are closely connected in structure and extremely similar in attributes into the same group.Finally,through the mining and analysis of the massive data on social network s,this paper designs and implements a group detection system based on user profiles.It forms an effective application system for collecting and analyzing characteristics of specific people and their groups.The results of testing and application show that the system has good performance and the function meets the requirements.