Construction Of Event Knowledge Graph For Militaryauxiliary Decision-Making

Military events,as an essential source of intelligence collection,record military activities of each country,and are an crucial basis for military decision-making.The introduction of event knowledge graph can effectively utilize military events and provide data support for decision-making tasks such as information aggregation and military question answering by integrating,correlating and expanding isolated and scattered events.Therefore,with analysis of three typical military decision-making demands including event semantic search,spacial-temporal target tracking and topic aggregation,this thesis constructed an event knowledge graph for military auxiliary decision-making by event knowledge representation,event knowledge fusion,and event topic knowledge extraction.Moreover,an auxiliary decision-making prototype system based on military events was designed and implemented,providing a valuable reference for constructing a general decision-making intelligence engine in the military field.The main research and contributions of this thesis are:（1）Based on typical military decision-making demands,the corresponding military auxiliary decision-making tasks and auxiliary functions supported by military event knowledge graph were described.In addition,according to military decision-making demands and characteristics of military events,a military event knowledge representation model was designed.Furthermore,a military event-centric knowledge graph containing key elements including type,time,space,combat objectives,and combat missions was constructed.（2）In order to better support decision-making demands including event semantic search and spatial-temporal target tracking,the knowledge fusion approaches of time element,space element and combat target element were respectively proposed based on domain knowledge from Baidu Encyclopedia and Wanguo Weapon Web.The approaches improved knowledge schema,eliminated entity redundancy and expanded relations and attributes of entities,realizing the multi-dimensional semantic correlations among events in time,space and combat objectives.Furthermore,the entity alignment and construction of co-reference relations among combat tasks were realized based on the ERNIE_Base-Linear semantic matching model,which provided support for event trend analysis.（3）In order to effectively recommend accurate event information,a method of event topic detection and extraction was proposed by text modeling technology and cluster analysis.Firstly,an event text modeling method with enhanced topic features,combined with LDA topic model,Bi-LSTM network,attention mechanism and Text-CNN,was proposed to alleviate the problem of poor text embedding effect caused by sparse semantic information and inconspicuous topic features of short texts.Secondly,topics of cross-type events and of same-type events were detected by multi-level text cluster analysis.Last,event topics were extracted by the combination of word cluster analysis and keyword ranking algorithm.（4）A military auxiliary decision-making prototype system was implemented based on the military event knowledge graph constructed by this thesis,which provided users with data support by functions including event semantic search,key information query,spatial-temporal correlation and topic aggregation.