Research And Application Of Emotional Analysis In Weibo Text Based On Deep Learning

With the improvement of technology and the development of internet technology,social media such as Weibo has become one of the main information sharing platforms in China.Most Weibo texts have a strong emotional color.Conducting emotional analysis and research on Weibo texts is beneficial for relevant government departments to regulate public opinion and enterprises to carry out crisis public relations.On the other hand,it is also beneficial for users to quickly obtain various information.Therefore,conducting emotional analysis research on Weibo texts has social significance and commercial value.Due to the diversity,obscurity,and lack of contextual information in Chinese Weibo texts,traditional sentiment analysis models cannot effectively handle the characteristics of Weibo texts.Therefore,this thesis focuses on this situation and the main research work can be divided into the following parts:（1）A Chinese Weibo text sentiment analysis model has been studied and proposed to address the issues of insufficient semantic understanding and limitations in feature extraction in traditional sentiment analysis models.The model first uses the Ro BERTawwm-ext model in the word embedding layer to convert text into dynamic word vectors;Secondly,in the feature extraction layer,the Bi LSTM model is used to perform deep emotional feature extraction on the word vector output from the word embedding layer;Then,in the multi head attention mechanism layer,highlight the semantic features of keywords and capture local features of input sequences;Finally,all the extracted emotional semantic features are connected in the output layer to complete text sentiment polarity classification,and the sentiment polarity category of the text is output.（2）A series of experiments were conducted on the emotional analysis model proposed above.In the ablation experiment,the advantages of the Ro BERTa-wwm-ext model in training word vectors and the role of the multi head attention mechanism in capturing local emotional features of the input sequence were verified;In the comparative experiment,the model presented in this paper was tested in weibo_senti₁00k dataset.The accuracy and F1 are higher than other models,verifying the feasibility and effectiveness of this model;In the parameter tuning experiment,the optimal parameter values of learning rate,Dropout,and Batch Size were optimized to improve the emotional analysis effect of the model on Chinese Weibo texts.（3）Based on the proposed sentiment analysis model,a brand topic sentiment trend monitoring system has been developed.Firstly,a requirement analysis was conducted on the system.Secondly,the system architecture and functional modules were constructed based on the requirement analysis.Finally,the system database was designed according to the division of functional modules.