Design And Implementation Of Event Causality Analysis System Based On Recurrent Neural Network

The event causality analysis task is to judge the causal relationship between the events that have occurred,analyze the degree of influence of an event on the occurrence of other events,and then mine the causal relationship between events to make event predictions.Because the traditional event reasoning and event prediction methods only target events in specific scenarios,their application scenarios are limited,it is difficult to generalize to general events,and the method itself is difficult to extend.Around this problem,this article introduces the concept of event information,designs an event causality analysis and event prediction model based on recurrent neural network,and adds an understanding of the dynamic and time-varying nature of causality in inference to strengthen the ability of event causal reasoning.The validity of the model was verified through the Sino-US trade war themed events,and a visualization system of event causality analysis was constructed based on the need for visualization of causality.The content of this article is as follows:1.Investigate multi-structured feature construction strategies for event data.The strategy aims to structure the event data and fuse more valuable information in the process of characterization.This article also encodes the event's semantic information into its feature vector,and combines multiple discrete features to achieve the diversity and characteristics of the event data It can improve the information richness of the event feature vector,enhance the network's understanding of the event itself,and improve the reasoning ability.2.Research on event causality analysis method based on recurrent neural network.Aiming at the time-varying and statistical nature of the causal relationship,the modeling advantage of recurrent neural networks for time series data is used to construct event information of event development,and the causal laws of events are dynamically encoded into time series information and stored in the hidden In the state,both statistics and dynamics are taken into consideration to realize the adaptive mining of causality between thematic events related to the Sino-US trade war.3.Research on event prediction related algorithms.Based on the results of causal analysis between events,research event prediction algorithms,make full use of the model's coded causality,and combine dynamic historical data to improve the rationalityof prediction results.The thematic events related to the Sino-US trade war are predicted through experiments,and compared with the real events,the effectiveness of the prediction algorithm is verified.4.Based on Qt5,an event causality analysis visualization system was designed and implemented.Combining with the actual application requirements of event causality analysis,the system software module is designed.The Qt5 platform is used to implement the graphical user interface and interaction logic,and to visualize the causal relationship of events and the prediction results of events.The Qt view system is used to fully and clearly show the complex causality,and through the multi-thread mechanism,the system can still quickly respond to user requests when making algorithm calls.This thesis validates the system's effectiveness and integrity by performing functional and performance tests on the system.The research results help to improve the efficiency of auxiliary decision-making and have a good application prospect.