Topic Classification And Risk Assessment Of Agricultural Public Opinion Based On Deep Learning

These information about the safety and price of agricultural products are closely related to the livelihood and health of the general public,it is easier to cause widespread concern and heated discussion on various network platforms.Therefore,it maybe has potential risks and hidden dangers to the healthy development of industry and social stability and harmony.Firstly,the paper made a comprehensive and in-depth study on the hot spot discovery and topic classification of agricultural network public opinion,constructed a topic popularity prediction model named LSTM – ATTN.Then,proposed an index system and a model of risk assessment based on the evolutional trend of network public opinion.Lastly,an empirical analysis is discussed according the "African Classical Swine Fever" event,which demonstrated effectiveness of the models in this paper.(1)In view of the large amount of agricultural public opinion information on various network platforms and various contents,LDA theme model is used to extract relatively independent hot topics from public opinion events.Then the LSTM-ATTN model with attention mechanism is constructed to realize the scientific prediction of the heat change of each topic.In the empirical analysis of public opinion events of "African Sine Fever",the maximum MAPE error between the predicted results and the actual data was only 11.3%,and the LSTM-ATTN model had higher prediction accuracy and faster convergence speed than the traditional LSTM model.The above research results could not only achieve more accurate topic classification and heat prediction of public opinion,but also laid a scientific quantitative evaluation basis for the subsequent evolution trend analysis and risk assessment of public opinion.(2)An improved model E-SEIR model was constructed to to realize the simulation of netizens’ emotion evolution and information diffusion in the process of agricultural network public opinion spreading.In the infectious disease model SEIR,the emotional evolution process of netizens was innovatively embedded,and the influence of individual emotional tendency was acted on the process of public opinion diffusion through the probability of information transmission.In addition,taking into account the actual situation such as the guidance and control strategy of functional departments,the rules of personal status transfer have been improved.The paper analyzed the influence of key factors such as netizens’ emotional tendency,key opinion leaders,and the distribution of initial emotional tendency based on E-SEIR.The above research provided a basis for the scientific research and judgment of the evolution and development trend of public opinion,and also provided scientific quantitative indicators for public opinion risk assessment.(3)Integrating the content of public opinion events,the attitudes of netizens and the evolution of public opinion and other important factors,a risk assessment index system for agricultural online public opinion has been constructed.The risk levels of public opinion events in different stages of development were reasonably divided through grey correlation analysis.The index system includes four first-level indicators,namely,the nature of public opinion events,netizens’ attention,netizens’ emotional tendency and the force of key opinion leaders.The index system covers all kinds of subjects and related influencing factors involved in public opinion events.Scientific risk assessment can be carried out by fully combining real-time statistical data of public opinion events and prediction data of future development trend,which can realize risk grading of different stages of public opinion development.Aiming at the actual needs of agricultural online public opinion research,the paper realized the classification of hot topics of online public opinion and the prediction of the trend of heat change.By improving the SEIR model,the modeling and simulation of the diffusion of public opinion information and the evolution of netizens’ emotions were realized.Combined with the real-time statistical data of public opinion and the evolution trend forecast data,the staged risk classification and scientific evaluation of public opinion events have been realized.The research in the paper has enriched the modeling and quantitative methods of agricultural network public opinion,which helps to reveal and discover the law of occurrence and development of public opinion,and provides a scientific basis for scientific risk assessment and risk prevention and control of public opinion.