Research On Emergency Extraction Based On Semantic Feature Fusion

In the era of big data,the explosive growth of network information not only speeds up the efficiency of information interaction,but also increases the difficulty of intelligent data analysis.How to obtain emergency information quickly and accurately from a large number of text corpus is of great significance to the release and early warning of emergency and to provide information basis for the customization of emergency strategies.At present,there are often two problems in the extraction of emergency events.First,in the extraction of event trigger words,the meanings expressed by trigger words in different contexts are inconsistent,which is ambiguous and easy to mislead the identification of event types of trigger words.Second,the deep semantic relationship between the arguments is not considered in the extraction of event arguments.This dissertation makes an in-depth analysis of the existing emergency extraction methods,and proposes an emergency extraction method based on semantic feature fusion for the extraction of event trigger words and event arguments.The specific research content is as follows.(1)Emergency trigger word extraction model based on semantic and contextual features fusion.Firstly,BERT pre-trained language model was used to represent the pre-processed word sequences,and dynamic contextual word vectors were obtained.Then,joint vectors were constructed by fusion with static word vectors represented by Word2 vec model to solve the ambiguity problem of trigger words.Then,the BiGRU network is used to extract the forward and backward semantic information of the text to obtain the contextual semantic features.Then,the graph attention network is introduced to extract semantic information,remove redundant semantic features,and extract semantic features from different semantic Spaces.Finally,CRF classifier is used to mark the results of weighted fusion features and complete the extraction of emergency trigger words.Finally,the F1 value of 74.72% was obtained,and a good extraction effect was achieved.(2)Extraction model of emergent arguments based on the fusion of semantic and dependency features.In this model,BERT word vector,POS tagging,named entity and BiGRU context semantic features were extracted and fused into multi-feature vectors to enrich the semantic features of event arguments.Then,the dependency syntactic tree of emergent events is constructed,and the graph attention network is used to extract the more important dependency syntactic information of the event arguments to obtain the dependency relationships among the event arguments in the sentence.Then,a gating mechanism is proposed to realize the dynamic fusion of contextual semantic features and the hidden representation of graph attention network by weighting,so as to preserve the effective features.Finally,CRF joint decoding is used for classification operation to complete the extraction of emergency arguments.The F1 value of 72.95%is obtained in the experiment.The results show that the model has a good effect on the extraction of emergency arguments.Figure [16] Table [8] Reference [67]...