Research On Sentiment Analysis Of Short Texts Based On BERT Algorith

With the rapid development of e-commerce and social networking platforms,Internet users can post their opinions on products they buy and social events on the Internet,thus generating a large amount of text data.These texts are often referred to as "short texts",which are characterised by strong topicality,real-time,wide dissemination,fast update,fragmentation,unconstrained wording and sparse features,making it difficult for traditional algorithms to obtain effective features from short texts.In this paper,we have developed an algorithm for sentiment analysis of short texts and deployed the corresponding algorithm model for verification:(1)The BERT-RCNN model was constructed for experiments on the short text sentiment classification benchmark datasets WEIBO and SMP2020,and the results showed that BERT-RCNN achieved better sentiment analysis performance compared to the selected other baseline models,and the BERT-RCNN model for short text sentiment analysis,the model was able to store more semantic environment information and extract more deeper text features.(2)A BERT-DPCNN model combining pre-trained model BERT and deep pyramidal convolutional neural network DPCNN was constructed for short textual sentiment analysis,which can extract not only sentencelevel textual dependencies but also word-level textual local dependencies,while alleviating the problem of gradient disappearance of deep neural networks.And the experimental validation is also performed on the selected sentiment classification benchmark dataset,and the results show that BERT-DPCNN achieves better performance than the selected baseline model.(3)This thesis designs and implements an automatic short text sentiment classification system,which accesses the classification interface provided by the short text sentiment analysis model deployed through the Flask framework,and can classify single short text sentiment messages uploaded by users,and can obtain user-posted content under microblog-related topics for batch sentiment text analysis through crawler technology,and present the analysis results to users after statistics.The results are then presented to the user.