Research On Social Network Spammer Detection Based On Semantic Information And Feature Selection

Social network platforms provide users with convenient communication and interactive tools,which can instantly share various content related to life and work,including text,pictures,and videos.With the advancement of computer technology,social networks have taken the world by storm.However,the large number of users on social network platforms and the convenient conditions for publishing content have attracted a large number of malicious users on the Internet.Although researchers have proposed many methods for detecting spam and spammer,there are still many problems,such as the huge workload of building a thesaurus,the ineffectiveness of word-based language features,and low detection accuracy and efficiency.In order to improve the situation,this thesis studies the detection models of social network platform spam and spammer.The main research content includes the following two aspects:1.Aiming at the problem of spammer disguising the text content and causing the language features of the malicious lexicon to be unavailable,a self-attention-based BiLSTM neural network model combined with the lightweight word vector model BERT(ALBERT)is proposed.The model first uses ALBERT to transform the social network text into word vectors,and then enters to the Bi-LSTM layer in the form of word vectors.After the feature extraction and the combination of self-attention layer information focus,the final feature vector is acquired.And finally,the feature vector is passed to the Soft Max classifier.Experiments show that our proposed model is superior to other models2.For the problem of the redundant data features caused by huge amount of data and the model perform not well,a model based on meta-heuristic algorithm for feature selection and classifier for spammer detection is proposed.First,the binary whale optimization algorithm is used to analyze the features and select some of the features that have the greatest impact on the detection results to form a feature subset,and then these features are used to train the classifier.Our proposed model can effectively identify spammer and non-spammer.Finally,to prove the superiority of the model,the model was compared with several existing sophisticated methods.The result shows that the accuracy of this model is higher than the other models.