Research On The Construction Of Causal Graph For Tactical Task

The tactical task is an essential part of military operational command.Summarizing past tactical tasks plays a vital role in guiding future military operations.A tactical task is a kind of dynamic event knowledge.The knowledge graph used to describe entity knowledge is difficult to describe dynamic and complex knowledge such as event-to-entity and event-to-event.The demand for intelligently acquiring knowledge is growing with the advancement of information technology.It is urgent for people to construct a knowledge carrier that describes the evolution law and development logic between events to explore knowledge.Therefore,the Event Logic Graph(ELG)arises at the historic moment.ELG is a logical knowledge base for describing the evolution law and pattern between events.This paper constructs an ELG for tactical tasks.Firstly,because of the lack of a special corpus for constructing ELG in the military field,tactical tasks’ event definition and event relationship are defined.Based on this,a certain scale of meta-task corpus and sub-task corpus is annotated.Then,tactical task event extraction and relation extraction are carried out based on the constructed corpus sets.The main research contents include:(1)Aiming at the problem of dense distribution of tactical task events and considerable context noise of trigger words,this paper first uses the BERT(Bidirectional Encoder Representation from Transformers)model to obtain the sentence vector with semantic features,and then uses Bi-LSTM(Bi-directional Long Short-Term Memory)to capture the sentence context features.Finally,CRF(Conditional Random Field)is used to learn the labeling rules,so as to extract event trigger words and arguments from the tactical task text.The results show that,compared to comparative tests,the suggested method in this research performs well in the event type identification and argument extraction of tactical tasks.(2)Aiming at the lack of display relation words in tactical task text,an event relation extraction model is proposed based on Self-Attention and Bi-LSTM.The model improves the accuracy of relation extraction by capturing the long-distance features of sentence context and deep semantic information.The experiment results show that the values of the model reach 71%(meta-task corpus set)and 70%(sub-task corpus set),respectively,which indicates the effectiveness of the model in realizing tactical task event relationship extraction.(3)Aiming at the problem that meta-task events have same semantics and different expressions,the improved K-means++algorithm is used for event generalization of meta-task events.This paper improves the K-means++algorithm from two aspects:1)the BERT model is used to replace the traditional TF-IDF method for event sentence vectorization;2)According to the clustering object for the event,the distance between the two sample points is measured by calculating the event similarity.Based on the results of the experiments,the enhanced Kmeans++method can improve the clustering effect,and the F value is increased by 14.2%.Based on the above research methods,this paper extracts the tactical task events and event relations of island landing operations,and determines the relationship between meta-task and sub-task using expert experience knowledge.Finally,the tactical task ELG is constructed,and this paper uses Neo4j to realize the ELG’s storage and visualization and analyzes the ELG’s application in mission planning.