Research Of “Water Army” Recongnition Based On DBN

With the development of information technology, cyberspace has been widely used in everywhere of human life as an important way for people to communicate and interact with each other. The distance between netizens has been largely shortened by the cyberspace. Information can be easily and quickly released, received and shared. Information in the cyberspace has a more and more huge influence on human’s life, as people are becoming more and more dependent on the cyberspace. It may affect how to think and how to make decisions. As the evolution of cyberspace someone called “Internet water army” has been generated among the internet users. The Internet water army(IWA) represents those who release and share lots of fake or spam information in the Internet on the purpose of gaining interest. These IWAs are always criticized for disrupting the correct order of the Internet. They are evolving with a pattern similar to real users in the long confrontation. Therefore, how to identify the IWA accurately and quickly has becoming a challenge to be addressed.In this paper, aimed at improving performance of the existing algorithms for water army recognition, we propose a deep belief network(DBN) based approach to identify Internet water army, using the most influential microblogging platform Sina Weibo as the data source and using real features of Weibo users as the basis. This approach not only combines user features and text features but also uses abnormal activities with time to tell features of IWA in time dimension. The experiment verifies that this approach has an effective performance in IWA recognition in practice.Considering the rapid development of cyberspace and the increasing volume of information which leads to the problem of time-consuming while processing data, some improvements in preprocessing and training procedures have been proposed to advance the performance of the approach. Parallel method was utilized to speed up of the training of the model. Experiments showed that these improvements can advance the efficiency of IWA recognition model, solving the problem of time-consuming of model training for massive data.