Research On The Account Classification Methods In Social Media

With the rapid development of Internet technology,social media has gradually become an important tool for people to exchange and share information.As the main body of people's activities on social media,accounts play an important role in transmission,dissemination and reception of information.Quickly and efficiently identifying accounts with specific value information and classifying management from a large number of social media accounts is conducive to establishing a more complete information management system in social media,and plays an important role in building a healthy and orderly social media environment.The social media account classification includes two parts: spam account filtering and account theme recognition.The purpose of the spam account filtering is to identify and filter out useless accounts with too little value information,such as marketing accounts,robot account,etc.the purpose of the account theme recognition is to distinguish accounts focusing on different theme information,such as politics,military,news,etc.The existing account classification method is not comprehensive,in-depth and accurate in network characteristic analysis and content feature analysis,resulting in poor effect of spam account filtering and account theme recognition.To this end,this thesis studies the classification methods of social media accounts,and the main contributions are summarized as follows:(1)This thesis proposes a method for spam account filtering based on user interaction behavior.This method mines user features from the interaction behaviors of content characteristics and network characteristics.Based on the analysis of account content characteristics,a plurality of short texts are combined into a long text as the overall account content information,the feature words are selected by means of information gain to construct a content feature vector.Based on the analysis of the network characteristics of the account,the user social relationship is used to construct a local network relationship diagram,and a large number of useful network features are extracted from both the user and user neighbors.In the comparative experiment,the effect of spam account filtering based on the joint features of content and network is obviously better than the single feature,the accuracy rate of spam account recognition reaches 91.8%.(2)This thesis proposes a method for account theme recognition based on convolutional neural network.The traditional account theme recognition is based on bag of words model,which not only ignores the semantic association information between text words,but also ignores the topic association information between the account texts.In this thesis,the word2 vec model is used to represent words as low-dimensional dense vectors,which has the advantage of maintaining the semantic correlation between adjacent words.Meanwhile,the topic probability distribution information of text is introduced,and the convolutional neural network is used to extract the local topic relevance features.By analyzing the experimental results of before and after the introduction of tweets topic distribution information,and comparing the experimental results of traditional bag of words model.The validity of semantic association information extraction and topic related information extraction is verified.