Analysis And Prediction Of Emergencies Based On Emotion Sequences

Against the background of "speed increase and fee reduction",the blowout development of the mobile Internet has not only increased the penetration rate of the Internet,but also provided a good platform for people to communicate with each other.This has made social networks a way for the masses to express their sentiment and public opinions.The development of social public opinion on the Internet has made the emergence and outbreak of public opinion on the Internet more complicated.In the supervision of emergency network public opinion,mining important information in social text is the focus of work.The important information includes the web users' emotional sentiment when commenting on emergencies,and related entities,entity attributes,and inter-entity relationships of the event:On the problem of word vector representation for text,this dessertation proposes a word vector representation method based on AME(Average Meta-embedding).This word vector model combines Word2 vec and Glove methods,which improves the performance of the text sentiment classification model to a certain extent.For the problem of text sentiment recognition,the dessertation proposes a sentiment analysis method based on sentiment dictionary and self-attention mechanism,which avoids the problem of low accuracy due to incomplete dictionary coverage and insufficient recognition of emotional network terms by deep learning networks.It makes full use of the advantages of the sentiment dictionary in the field of emotional words and the advantages of deep neural network models in feature extraction,reducing the limitations of the single model itself.For the analysis and forecasting of emergencies,this dessertation proposes a method based on knowledge map to analyze the components and characteristics of emergencies.From the two main quantitative indicators of temporality and emotional orientation as the starting point,find out the event-related topics through hierarchical clustering,and use the Bi-LSTM-CRF model to extract entities and entity relationships in social text to build a knowledge map of events.The form provides an effective reference example for further understanding the origin and trend of the event.Through qualitative and quantitative comparison experiments,the experimental results prove that the knowledge graph-based method proposed in the dessertation to analyze emergencies and the sentiment analysis method based on sentiment dictionary and self-attention mechanism is feasible.