Knowledge Base Construction For Internet Emergency Events

Emergency events threaten the security and stability of the society with its sudden and destructive moment.Public opinion of emergency events tests the emergency events management ability and governance level of relevant departments.In today's society,the Internet gives everyone the right to speak and becomes the main position of public opinion.This makes the Internet contain a lot of valuable emergency events related knowledge,but this strategic resource has not been effectively mined and utilized.As a tool of knowledge management,knowledge base can realize the efficient management of massive knowledge and effectively alleviate the contradiction between "massive information" and "knowledge shortage".Most of the existing knowledge bases are oriented to the general open domain,while the knowledge bases for specific domains need to be specially constructed.This thesis aims at the field of emergency events,constructs the Internet emergency events knowledge base,and provides a new idea for the mining,storage and management of emergency events related knowledge.The main research contents are as follows:(1)Emergency events data collection and processing.With the help of Python web crawler,selenium automation tool and Scrapy framework,the emergency events information of Baidu Sina Weibo and Post Bar social network platform is collected to form a preliminary event information base.At the same time,a series of preprocessing operations such as data cleaning,Chinese word segmentation,stop word filtering are carried out for the collected source data.In order to select high-quality data,a data selection strategy based on event sentence is developed.The involved objects in the emergency events is annotated with BIO annotation system,and an annotation data set including two types of entities: personnel involved(PER),organization involved(ORG)and non-entity(O)is formed.(2)Recognition of involved objects in emergency events.In order to identify the personnel involved and organization involved in emergency events,the BiLSTM-CRF model based on word2 vec word vector is trained.Experiments show that compared with the random initialization word vector,the embedded word vector improves the recognition effect of the model.At the same time,compared with the single HMM,CRF and BiLSTM models,the overall recognition rate of BiLSTM-CRF model based on word2 vec word vector is also higher.This model has a good effect on the recognition of involved objects in emergency events.(3)Emergency events knowledge base construction and the visualization system implementation.The emergency events knowledge base consists of two sub databases: the event information base and the involved objects base.The event information base is the basic database,which mainly includes the event related knowledge collected by the web crawler.The involved objects base is an extended database,which mainly contains the information of the involved object extracted by the algorithm.In order to improve the efficiency of knowledge acquisition,an Internet emergency events knowledge base system is built,which shows the basic information,public opinion information,the involved objects and other knowledge modules.Through the knowledge management function of each module,the maintenance and optimization of knowledge base is realized.This thesis collects the emergency events data on the Internet and carries out a series of data preprocessing operations,trains the model to identify the involved objects in the emergency events,and constructs the involved objects relationship diagram by human intervention,finally constructs the emergency events knowledge base including the event information base and the involved objects base,and developments the Internet emergency events knowledge base system.