Research And Application Of Event Detection Based On Semantic Information Fusion

Event detection aims to extract trigger words and identify event types in unstructured text in an automatic way,which is one of the research focuses of information extraction task at present.Event detection is the key task of constructing the theory graph and realizing the application of knowledge reasoning and question answering system.Its accuracy greatly affects the performance of the downstream task.However,the existing event detection models mainly enhance the feature representation of word vectors by integrating contextual word vector features and syntactic information,and lack the use of other valuable semantic information.based on the analysis of Semantic Information produced by other subtasks in the field of information extraction,this paper proposes an Event Detection model based on Semantic Information Fusion(EDSIF).By integrating entity type,event element and dependency syntax,the accuracy of event detection task is improved.Firstly,the relational graph is generated based on the semantic information of relational class,and the spatial information contained in the adjacency matrix is captured by multi-scale convolutional neural network and fused with word embedding.Secondly,non-relational semantic information such as entity type and event elements is further integrated in the initial word embedding,which is encoded by the Bi-LSTM model to aggregate contextual information.At the same time,attention mechanism is used to guide the encoded word embedding representation.Finally,a Gated-GCN model is constructed to dynamically aggregate relational semantic information between adjacent word embeddings to enhance the representation ability of gated word embeddings.In addition,the Interactive learning framework for Event Detection and Argument Extraction(ILF-EDAE)is further proposed.By capturing the correlation between event detection and event element extraction tasks,an interactive learning channel is constructed between the two tasks,and the performance of the two tasks,especially the event detection task,is improved.Based on the ACE05 benchmark data set,the experimental comparison and analysis with the existing mainstream event detection models show that the F1 value of the proposed method reaches 77.6%,effectively improving the performance of event detection tasks.Finally,based on the existing theoretical research results,this paper combines theory with practice to design and implement an event detection system based on semantic information fusion.Based on Flask web application framework,the system carries out information exchange on the front and back ends,and realizes the visualization of event detection results through model deployment,data preprocessing,event detection and other processes.