Developing A Identifying System For Multi-dimensional Network Communities Based On Transfer Learning

As the rapid development of network applications, with users’ different quality,public opinions on unexpected events has a major impact on people’s life and work. Andsome extreme or even seriously events affect public opinions, social security and stability.In order to prevent the adverse effects of maintaining overall social stability, peoplerequire a comprehensive grasp of public opinions, the development of events and trends,identification of public opinion in the evolution of opinion leaders and other specialpopulations. In this paper, users’ speech at the forum and behaviors of Internet users inthe network, using the application of knowledge of relational data migration method tosolve the problem of multi-dimensional public opinion, provides key technologies to theunderstanding of public opinion.First in this paper, the purpose and significance of multi-dimensional understandingof public opinion and opinion leaders is described, and methods of identifying networkopinion leaders based on relational data are given explicitly, and comparisions ofaccuracy evaluated from different methods for identifying network opinion leaders. Thispaper also briefly introduced other key related technologies used in opinion leaders’recognition.This paper presents three technical points: a multi-dimensional recognitiontechnology in a single-source domain, migration technology for single-source domain tothe target domain, and migration technology for multi-source domain to the targetdomain, and designs a four modules prototype system. The system is able to overcomethe lack of target domain labels, shortcomings of the traditional method of isolated tasks,using the relational model to consider a number of issues simultaneously, which gives amore comprehensive understanding of public opinion.Finally, the paper presents experimental resuts evaluated from the implementedprototype based on the model proposed, and the analysis of the experimental results,which demonstrates the accuracy and robustness of our recognition system on thenetwork public opinion compared to other methods.