Research On China-ASEAN Weibo Sentiment Analysis Based On Deep Learning

With the continuous deepening of China-ASEAN interconnection,research on China and ASEAN countries’ network public opinion is particularly important.There is relatively little research on China-ASEAN folk opinion analysis.Analysis and research have strong theoretical significance and practical value.First,the Weibo review data related to the "ASEAN" topic was collected and selected as the experimental data set of this article.Based on the characteristics of Weibo data and the needs of the public opinion observation database,the sentiment analysis algorithms of CONV-SAWB and LST-SAWB are proposed.Conduct hyperparameter tuning experiments and network model training,use the trained model to predict and classify the test set,and the comparison of experimental results shows that the evaluation indicators of CONV-SAWB and LST-SAWB are better than SVM,indicating that the deep learning model Traditional machine learning models can better perform sentiment analysis of China-ASEAN microblogs.Secondly,since CNN only performs local feature extraction,LSTM is computationally intensive and the contextual connection to long text will be weakened.The study combines the two algorithms,and proposes the CLST-FMSA fusion algorithm,which uses CNN to extract feature features to reduce the amount of data calculation and the learning and memory ability of LSTM to better classify emotions.Although LSTM has a long-term memory function,it is still not possible to obtain the information behind the sentence.It is not comprehensive to rely on the previous historical information.Therefore,the LSTM network is improved,and the CBLSFMSA fusion algorithm is proposed so that it can integrate the previous information and the following input sequence to analyze the emotional polarity as a whole.Considering that the information contained in some special sentiment words will have a key impact on sentiment classification,the Att-BLS-FMSA fusion algorithm is proposed to introduce the attention mechanism to the two-way long-short neural network,and increase the weight of important parts in the semantic features to generate the final Text features.An experimental comparative analysis shows that multi-model fusion makes prediction of emotion classification more accurately than single model.Among them,Att-BLS-FMSA fusion algorithm has more advantages in emotion classification than other algorithms.Finally,the multi-model fusion proposed in this paper is applied to the sentiment analysis of the China-ASEAN marine big data platform,and users can query the sentiment polarity and scores of various indicators in ASEAN countries.It not only meets the needs of China-ASEAN marine big data platform for private public opinion observation,but also provides strong data support for public opinion surveys and public opinion control,and provides decision support for the government to analyze bilateral relations and predict risks,which fully reflects the actual value of the research results of this paper.