Causal Relationship Recognition For Emergency Events

The rapid development of information networks brings hundreds of millions of events and topical content to the shared platform every day.Events are an important type of expression.Event-oriented research has attracted more and more people's attention.Intrinsic connection of nature,this internal connection reflects the various deep semantic relations between events,and the most important relationship is event causality.Event causality recognition is expected to provide technical support for event-oriented knowledge representation,information retrieval and automatic question answering systems in the field of natural language processing,and has important research significance for the semantic understanding of emergency texts.At present,methods for identifying causal relationships in events mostly focus on pattern matching and statistical machine learning methods.The method based on pattern matching has better effect on the display of causal relationship with display related words,but the recognition effect corresponding to implicit causality is poor.The method is not universal.The statistical-based machine learning method needs to pre-label the training corpus,and has certain dependence on the features.Generally,the recognition effect is better as the corpus size increases.Aiming at the above problems,this dissertation proposes an LSP-based event causality recognition method.This method adds syntactic pattern matching on the basis of single event matching to enhance the recognition effect of causal event pairs and reduces the dependence on text corpus size.The main content and innovations include:(1)Event homomorphism based on maximum entropy modelIn the event causality recognition process,the same event trigger word may be misjudged in the causal event pair because it does not point to the same entity.At this time,it is often necessary to discriminate the same trigger word to help improve the event.Performance during the extraction process.Aiming at the high density of news files in emergencies and the similarity of news language styles in the same category,an event-based finger-diffusion method based on the maximum entropy model is proposed.Firstly,the artificially labeled identical finger event object is used as a positive and negative example of the same finger digestion,respectively training the weight values of the positive and negative examples,and analyzing the event elements according to the characteristics of the event,extracting semantics,distance,keywords for the document.The characteristics of the other aspects;then,using different feature combinations to cycle iteration,training the maximum entropy model;finally,using this model to calculate the pair of events to be resolved in the test corpus to complete the event with the same finger digestion.(2)LSP-based event causality identificationBased on the event-finger-dissolution of the maximum entropy model,a method based on LSP event causality recognition is proposed.Firstly,based on the CEC2.0 corpus,by combining event trigger words,event elements and related words,starting from the text semantics and syntactic structure,the LSP pattern is extracted from the training corpus by using the existing annotation information of CEC corpus;then the vocabulary is extracted by statistical methods.The syntactic pattern is combined with the commonality of causality to calculate the correlation strength to obtain the LSP candidate set.Finally,the pattern matching is performed in the test corpus through the candidate set,and the matched event pairs are calculated,so that the event pairs greater than the threshold are causal events.Yes,in order to achieve the identification of the causal relationship of the emergency.Experiments show that the LSP-based event causality recognition method has a significant improvement in recognition efficiency compared with other methods.Figure[7]Table[11]Reference[52]...