Research On Trend Prediction Of Mega-event Using Massive News Data

The trend prediction of mega-event refers to the prediction of the evolution of current affairs or military events affecting the peace and stability of a country or region,and is a hot issue in the field of international relations research.The advent of the era of big data and the development of artificial intelligence technology make it possible to predict the trend of mega-event based on public news data.The quantitative thought of "event data analysis method" in the field of international relations research was adopted in this paper,and the trend prediction of North Korean nuclear behavior and the detection and prediction of symptom events in the South China Sea dispute were carried out to meet the demands of existing research methods for the construction of characteristic indicators and the causal traceability of event trend.The special news data of massive events were obtained based on the web crawler technology,and processed using natural language processing(NLP),machine learning and other technologies to construct a prediction model and analyze the experimental results.The main contents of the paper were as follows:Firstly,since the trend prediction method of current mega-event based on massive news data depends heavily on experts' knowledge in the construction of feature indicators,the universality and timeliness of related methods have been greatly restricted.To solve this problem,a trend prediction method of mega-event was proposed in this paper combining semantics and event characteristics.The Latent Dirichlet Allocation(LDA)model and the event extraction technique based on pattern matching rules were used to automate the construction of related feature indicators from the two aspects of semantics and events.An improved model IDFLDA was presented to handle the problem that LDA had bias in feature word extraction.The complementary characteristics of the two types of characteristic indicators was experimentally analyzed to demonstrate the superiority of the fusion features compared with the single features.Using the North Korean nuclear behavior trend prediction as an example,the best prediction accuracy of the proposed method for the full time period reached 86.2%,which was better than the traditional method based on experts' knowledge to construct the feature index.Secondly,the detection and prediction methods of symptom events affecting the development trend of the South China Sea dispute were studied in order to explain reasonably the trend prediction results on mega-event in this paper.Considering the detection problem of symptom events as a Named Entity Recognition(NER)task,a detection model of symptom events was proposed based on Bi-LSTM-CNN-CRF;Taking the proposed detection model as the data annotation module,a prediction method of symptom events was designed according to LDA and multi-label logistic regression;The detection and prediction model of symptom events was built on the special news data of the South China Sea dispute,and the experiments were carried out to verify the presented methods.Finally,the main work and further research of the thesis is summarized.