Research On Event Detection Based On Sentence Type Information

Event detection is an important sub task in the field of information extraction,and its main purpose is to identify trigger words and classify trigger words in sentences according to the context.It is of great significance for the development of automatic information extraction technology.The diversity of event information in natural language brings great challenges to this task.This article focus on the event type information of sentences,the main research contents are as follows:(1)Event detection method via sentence type clusteringAiming at the problem that most of the current event detection research ignores the event correlation between sentences,this article proposes an event detection method based on sentence type clustering from the perspective of event clustering.Through the event clustering model based on semi-supervised learning,we can construct the correlation between sentences and obtain the sentence clustering type features.Experiments show that after incorporating the sentence type features of event clustering into the event detection model,the model has improved on the ACE Chinese and English datasets.(2)Event detection method via sentence type classifyingAiming at the previous problems,this article proposes an event detection method based on sentence type classification from the perspective of event classification.Through the event classification model based on BERT and attention mechanism,we can obtain the sentence classification type features.In order to further utilize the text information of the event type label,this article adopts the machine reading comprehension model to fuse event type features at sentence granularity,and combines the contrastive learning module to shorten the distance between trigger words belong to the same type.Experiments show that this model has greatly improved on the ACE Chinese and English datasets compared to the benchmark model.(3)Event detection via sentence hierarchical event typeAiming at the problem that most of the current event detection research ignores the hierarchical event correlations among triggers,this article proposes an event detection method via sentence hierarchical event type.We construct trigger graphs based on the correlations of event type and superordinate type respectively among triggers in the training set.Based on the hierarchical event type information of the sentence,,the candidates are associated with the nodes in the graphs,and the event type features are fused at the word granularity.Combined with dynamic multi-pooling features of candidates,trigger words are classified.Compared with the benchmark model,the model has significantly improved on the ACE Chinese and English datasets.This article obtained the event type features by event clustering and event classification respectively for event detection.As the sentence classification type features are more specific,this article starts from the sentence granularity and word granularity respectively,to learn the event correlations among sentences and triggers.Finally,experiments indicate that the importance of sentence type features for event detection.