Event Evolution Graph Construction In Political Field

With the support of big data and the rapid development of natural language processing technology,the demand for interdisciplinary integration has surged,and social computing has become a priority research area.As a branch of traditional social science,political science mainly studies political phenomena and political relations in human social life in order to reveal the nature and regularity of the process of political phenomena.At the same time,the big data era has made the scale of corpus in the political field expand rapidly and the corpus more available,so the demand for automatic text processing and event relation network analysis in political science is increasing.The relationship graph structure constructed by automatically extracting information from large-scale data can not only efficiently complete the knowledge discovery task and clarify the correlation between events,but also facilitate researchers to understand better and grasp the development trend of events,which is beneficial to promote relevant research in the political field.So,this paper mainly studies the construction of the event evolution graph in the political field.Event Evolution Graph is a logistical knowledge base that reveals the evolutionary rules and patterns among events,which can be used for the logical analysis of political phenomena and behaviors.This paper proposes a set of framework to construct the event evolution graph in political domain,and uses natural language processing technology to complete the automatic extraction of event information and event relationships,which greatly reduces the workload of manual information extraction,and lays a good foundation for the construction of subsequent event evolution graphs in this field.In the information extraction task,this paper chooses a more effective supervised learning method,while the high-quality domain tagging data is very scarce and the data labeling template is missing.In view of above problems,this paper has developed a set of political domain data labeling schemes and built a PEG database that provides data support for event extraction and relationship extraction methods.For the event extraction task,this paper proposes a pipeline model of a dynamic multi-pooling convolutional neural network based on word vector correction and a joint model based on a pre-trained model.The experimental results show that both methods can effectively improve the effect of event extraction compared with the baseline model dynamic multipooling method.The dynamic multi-pooling convolutional neural network based on word vector correction improves by 17% and 3% respectively on F1 value,and the method based on pre-training model improves by 41% and 11% respectively on F1 value.For the task of relationship extraction between events,this article divides it into two parts: explicit relationship extraction and implicit relationship extraction,which use template matching and Bi-direction Long Short-Term Memory Networks incorporating the Attention mechanism respectively.The two methods can complement each other in the task of relation extraction.After the information extraction,in this paper,different databases are used to store the acquired data and the visual class library is used to generate and visualize the Event Evolution Graph.Therefore,the construction of the whole political field is completed.Based on the method of constructing Event Evolution Graph in the political domain,we can obtain the graph with great scale and high quality,which can assist the reasoning and prediction of political phenomena and behaviors.Meanwhile,the construction process also provides new ideas for the construction of event evolution graph in other specific domains.