Search Service Platform Based On Event Knowledge Graph

With the continuous development of the contemporary search service industry,along with the in-depth integration of search engines such as Google and Baidu with knowledge graphs,the development of key technologies such as deep learning and natural language processing,and the explosive growth of information,how can we start from massive sudden changes? It is a problem to dig out useful information from the incident data.The knowledge graph uses entities as vertices and the associations between entities as edges to describe static knowledge,but knowledge in the real world is dynamic,and emergency news is the main carrier of dynamic records of emergency security events.This paper is oriented to emergencies,based on the CEC Chinese emergencies corpus,designs and implements a search service platform based on event knowledge graphs,which can transform emergencies text into graphs to describe emergencies in the real world.The purpose of the relationship between the details and the event is to provide platform support for the study of useful information in emergency data,to provide decision-making assistance for the emergency response of the national emergency management department,public security and fire protection departments,and to provide past data for emergency warning Analytical techniques and assistance in future predictions.This platform consists of three sub-platforms: data collection platform,data labeling platform,event search and map construction platform.The data collection platform uses multiple threads to collect news texts related to emergencies from the Internet.The data tagging platform is mainly to perform manual and multi-rule matching on raw corpus,and to achieve data tagging by combining the BERT-Bi LSTM-CRF model.The event search and graph construction platform mainly uses the BERT-Bi LSTM-CRF model to transform unstructured emergency news text into structured,from text form to graph form,to realize the application of event search and graph visualization.The main research contents are as follows:(1)Data collection platform.This paper designs and implements a Scrapy-based multithreaded crawler framework for data collection,adopts a data cleaning method based on regular expressions,and proposes an emergency text filtering algorithm based on Doc2 vec.Filtering of irrelevant text.It mainly solves the problems of slow data collection and irrelevant content,and realizes efficient and accurate automatic data acquisition.(2)The data annotation platform mainly provides annotation services for raw corpus in the field of emergencies.This paper proposes a multi-mode emergency event tagging algorithm,which can reduce the cost of manual tagging in the process of part-of-speech tagging.For local data,a fast tagging algorithm based on multi-rule matching is proposed,and for global data,a fast tagging algorithm based on the BERT-Bi LSTM-CRF model is proposed.It mainly solves the problems of slow labeling speed,high labor cost and inaccuracy,and realizes fast and accurate intelligent labeling of data.(3)In the event search and map construction platform,mainly event extraction and map construction.In event recognition,an event recognition algorithm based on trigger words is used;in event element extraction,an emergent event element extraction algorithm based on BERT-Bi LSTM-CRF is proposed;in event relationship recognition,an event causality recognition algorithm based on pattern matching is adopted;At the application layer,the neo4 j graph database and the front-end are combined to realize event search and graph visualization.Mainly solve the problem of incomplete and inaccurate extraction of emergencies,and provide users with a friendly and interactive search interface.