Semantic Classification Based On Deep Learning And Its Application In Rumor Detection

In the era of big data,the Internet is full of various types of data information such as images,voice,and text,and text information plays an important role.Classification of topics based on the semantic information of text is an important subject in the field of text mining and big data.It is convenient for people to organize and analyze data.Therefore,semantic classification technology is of great significance.With the rapid development of the Internet,the spread of rumors has increased in speed and breadth,causing a phenomenon of "rumor flooding",which seriously affects people's production life and social order.Therefore,it is also very important to detect rumors.Through in-depth research on the currently popular networks such as Text CNN ?Bi LSTM?RCNN and attention mechanism,this paper finds that Bi LSTM and Text CNN networks have achieved good application results in the field of semantic classification.The combination of these two networks can give full play to CNN's feature extraction capabilities and Bi LSTM's context-dependent ability,but the disadvantage is that it does not reflect the importance of each word in the text,it cannot focus on important words.In order to solve this problem,a text semantic classification method based on Bi LSTM-Attention-CNN combined neural network is proposed.The attention mechanism(Attention)is added after the Bi LSTM layer to extract the attention score of the output information;after the attention layer,the k-max pooling layer is connected to extract the top k important words to enhance the expression ability of model features.Finally,Text CNN network is connected to give full play to its advantages of feature extraction and output classification results.The experiment was conducted on the data set of DBPedia ?AGNews and Sogou.The experimental results show that the model proposed in this paper works best when k is set to 8 in the k-max pooling layer.Compared with other network models,the classification accuracy of the Bi LSTM-Attention-CNN combined neural network model is improved by 1?2 percentage points.This paper proposes a Bi LSTM-Attention-CNN combined neural network structure and then applies it to the field of rumor detection.This model is used to subdivide the rumor text into 10 topic categories,such as social rumors,political rumors,and entertainment rumors.According to the research,every kind of rumor has its own characteristics,therefore,,rumor detection is carried out under each category of rumor text after classification.The rumor detection method used in this paper is BERT pre-trainedmodel.The experiment was conducted on the MCG-FNews of the Chinese Academy of Sciences' Internet false news data set.The experimental results show that the detection accuracy rate after classification is improved to a certain extent than before.In conclusion,this paper first conducted in-depth research on the current popular network models,and finds that the advantages of different networks have not been integrated.Therefore,this paper proposes the Bi LSTM-Attention-CNN combined neural network semantic classification model,and applies it to the field of rumor detection.The rumor data set is divided into different subject types of rumor,and the BERT model is used to detect under different types of rumors to improve the detection effect.