Design And Implementation Of Event Knowledge Graph Platform

Knowledge graph is based on the entity as the vertex,and the association between the entities as the edge,describing static knowledge,but knowledge is dynamically in real world.Narrative is the main carrier for recording dynamic knowledge.Aiming at the knowledge extraction of narrative,this thesis designs and implements an event knowledge graph platform,which can transform knowledge from unstructured text form into event-centric graph form to describe the relationship between things in real world.The event knowledge graph platform is composed of data acquisition platform,graph construction platform,parallel computing platform and labeling platform.The data collection platform is responsible for obtaining news texts related to specified events from the Internet,tracking hotspot events,and automatically updating data.The graph construction platform converts event information from textual form to graph form.The parallel computing platform provides parallel computing capabilities to carry the computational tasks in the graph construction process.The labeling platform provides the marker and industry experts with a rulemaking and sample labeling platform to ensure that the platform can adapt to different business scenario.Workflow of the platform is divided into text collection,event extraction and fusion,event knowledge graph construction and visualization.Firstly,the event-related text is obtained by data acquisition platform,and then natural language processing technology is used to analyze the text,extract event information,then organize and fuse event information,finally organize event into graph form,complete the event knowledge graph construction,and realize the transformation of knowledge into an event-centric form of the graph.The research results include the following four points:(1)Aiming at the problem that sometimes search engine inaccuracy may lead to irrelevant text in collection data set,an event text filtering algorithm based on entity co-occurrence is proposed.The entity information in the text is used to quantify the association between chapters and then according to the inter-text association network filters event text.(2)In the event fusion process,the event element formatting scheme is proposed for the inconsistency of expression due to the flexibility of Chinese expression.Anevent element formatting scheme is proposed to eliminate the impact caused by different expressions.A meta-event fusion algorithm based on event similarity is proposed to remove duplicate events and reduce redundancy.(3)In the construction of event knowledge graph,the event knowledge graph hierarchy is defined,and the event information is organized into a structured and hierarchical graph form according to the event category of the business scene in which the event is located.(4)In order to satisfy the storage and inquire of large-scale event knowledge graph,a special graph database is designed and implemented which can save the graph in a distributed environment and encapsulate the corresponding inquire interface to implement the event as input.The graph-based inquire,combined with a parallel computing platform,provides data input sources and storage addresses for parallel computing.Finally,the front-end technology is combined to realize graph visualization.After testing,the platform implemented in this thesis is able to construct event knowledge graph.