Research On Event Extraction And Intention Recognition Technology In Specific Field

In the era of rapid development of science and technology,it is necessary that people integrate existing scene information and make corresponding decisions.In reality,the event is expressed as a large number of domain text description information.It is an important research task for Chinese event extraction to depict the event portrait quickly and accurately based on massive text and organize a logical and clear event context.Compared with entity relation extraction,event extraction and intention recognition could represent information more intuitively and comprehensively,and have broad application prospects in real-world fields such as information retrieval,intelligent question answering,recommendation system,and event reasoning.Based on the analysis of domain-specific data characteristics,combined with the latest developments in natural language processing technology,this thesis proposes corresponding solutions to some of the problems faced by event analysis tasks,mainly including four aspects:(1)Aiming at the problem of low accuracy of Chinese meta-event extraction,this thesis conducts meta-event recognition based on feature analysis and dependency syntactic parsing,builds and expands the trigger vocabulary of 12 types of events,defines extraction templates for different types of meta-events,and proposes an activity recognition extraction model based on pattern matching and semantic dependence analysis;(2)Aiming at the problem of inconsistency in the representation of events in the Chinese domain and unintuitive presentation,this thesis defines the domain theme event framework,designs a four-layer theme event structure,and focuses on sensitive argument phrases in event information sentences and proposes the ERNIE topic clustering model with the introduction of attention mechanism;(3)Aiming at the problem of sparse domain data features and difficulty in organizing and relating event relations,this thesis proposes a domain external knowledge integration model based on target fusion,which enriches the event attribute features,and further analyzes the extraction mode of Chinese temporal relations and causal relations.For the extraction of explicit temporal relations and explicit causal relations,a relation extraction method based on fuzzy matching is proposed.For the problem that fuzzy matching is difficult to identify implicit relations,a scheme of relation recognition based on confidence calculation is proposed;(4)In order to realize the intention analysis and recognition of the target activity,this thesis constructs an intention network based on the meta-event set and the correlation between the events,and proposes an intention recognition model based on ordering graph.Based on single-target and multi-target feature analysis,the attention mechanism is introduced for vector representation of event argument features considering trigger words,time,space,and target attributes.In addition,effective intention recognition results are obtained based on semantic analysis and computational reasoning.Finally,the proposed model method is validated on the domain Chinese data corpus.The experimental process is divided into four modules:activity meta-event extraction,subject event extraction,event correlation relation extraction and event intention identification.Accuracy,recall rate and F₁-value are selected as evaluation indicators,and compared with the current popular research methods.The experimental results show that the Chinese specific domain event extraction and intention recognition schemes proposed in this thesis have achieved good results,and the accuracy and recall rates have been greatly improved compared with other methods.At the same time,its usability and practicality have certain advantages in domain data.