Research On The Model Of Group Public Opinion And Sentiment Analysis Under Sudden-onset Disasters

The explosive development of mobile media and instant messaging has made online social platforms the mainstream public opinion arena.However,the openness of online social platforms makes it difficult to analyze and guide group opinions,especially in the event of urban emergencies.The public’s ability to empathize with disaster information is high but their ability to discern is low.The widespread dissemination of false rumors and negative extremist remarks can have serious impacts on disaster relief and social emotions.Therefore,it is necessary to use modern technological means to analyze and predict public opinion in a timely manner.However,existing models and methods still have limitations in terms of practicality,and it is necessary to construct more comprehensive models to guide the practice of managing public opinion during emergencies.Based on existing research,this study focuses on textual data generated during realtime emergency disaster events and constructs machine learning models for analyzing public opinion and sentiment within the affected population.The main contributions of this thesis are as follows:(I)The Talbert opinion topic analysis model is constructed based on language pre-training models and unsupervised clustering algorithms.Initially,the ALBERT model serves as the backbone network,enhancing the contextual language context and overall semantic expression capabilities of the topic model.Subsequently,the topic information from the LDA model is integrated to address the issues of polysemy and poor representation of out-of-vocabulary terms.The fused data is then trained using the unsupervised clustering model Mini Batch Kmeans++ while employing batch processing techniques to improve the speed of topic inference on large-scale disaster texts.Ultimately,the Talbert model is established.(II)The Athlete opinion sentiment analysis model is constructed by integrating language pre-training models and fused topic knowledge.Similarly,the ALBERT language pre-training model is used for text embedding.Text CNN and Hatt are used to reduce dimensionality and recombine text features,thereby improving training speed while maintaining accuracy.The results obtained from the Talbert model constructed in this study are incorporated as external knowledge features into the sentiment analysis model,addressing the issue of the model’s poor perception of implicit sentiment.The final outcome is the construction of the Athlete model,which incorporates fused topic knowledge.Additionally,the UMAP algorithm is used to systematically analyze the underlying principles of incorporating topic knowledge in enhancing the effectiveness of the Athlete model from a visualization perspective.(III)Case studies are conducted to analyze the opinion topics and sentiments of various groups in the context of emergency disasters.Using the constructed Talbert and Athlete analysis models,the extreme rainfall disaster in Henan Province on July 20,2021,is taken as an example.This study analyzes the opinion topics and sentiment of four groups: the general public,social media,official media,and enterprises & institutions,from temporal and spatial perspectives.The feasibility and effectiveness of the proposed models are verified,providing scientific guidance strategies for managing public opinion during emergency disasters.Comparative analysis and case study results demonstrate that the proposed Talbert and Athlete models outperform previous state-of-the-art models,exhibiting higher practical value and suitability for group opinion analysis tasks in emergency disaster events.Government-related entities can utilize the model results to swiftly identify and respond to negative public opinions arising from disasters,thereby enhancing the emergency disaster public opinion response capabilities of management departments.