Research On Key Technologies Of Semantic Robot For Campus Network Public Opinion

Internet applications are very popular among college students,almost every student is using internet applications,and college students have a lot of spare time and energy to express their views on online social networking platforms,produced a large number of public opinion text.To excavate the emotional tendency of college students from mass college public opinion texts can help college administrators to prevent the occurrence of vicious public opinion events effectively,so the campus network public opinion semantic robot has been widely concerned.At present,with the development of word embedding technology and the development of deep learning classification model,the accuracy of emotion classification has been further improved.However,the existing network sentiment classification model is difficult to deal with the short text features are not obvious,resulting in unsatisfactory classification results.On the other hand,online public opinion platform is often multi-people interactive platform,the traditional model is difficult to deal with multi-people interactive emotional classification.Therefore,the research is carried out from the following aspects:(1)introducing Bert model into the task of sentiment classification on campus network,Bert model can discover the deep semantic information of sentiment text through the pre-training of domain corpus.(2)to solve the problem that Bert Basic classification model is not ideal for short text classification,a BERT-Pooling sentiment classification model is proposed,and the feature matrix generated by Bert language model is extracted and enhanced by Pooling method,to solve the problem that short text features are not obvious.(3)in order to solve the problem that the text information can not be considered in the dialogue scene,the text dialogue is transformed into a graph structure,which is used to process the relevant information of the target text.(4)in order to deal with the text dialog graph effectively,this paper uses graphgraph neural network to transfer message and update the text representation of the target node,so that the target node can get rich semantic information,improving the effect of text sentiment classification in dialogue.In order to verify the validity of the model,some comparative experiments are done in this paper.The experimental results show that the BERT-Pooling model proposed in this paper can improve the accuracy of sentiment classification task,and the CNN-Graph Sage model proposed in this paper can also achieve good results in sentiment classification task,compared with the dialogue text as a single text,compared with the traditional model of dialogue text processing also has a good improvement.