Research On Event Extraction Method Based On Question And Answer Pattern

With the explosive growth of data volume in the information age,the way of information dissemination has been rapidly changed,and easy access to information has brought about some information redundancy and overload at the same time.Event extraction tasks can quickly extract valuable information and present it in a structured manner,thus becoming a hot topic of current research.Most traditional event extraction methods focus on the design of complex neural network models and rely on a large amount of labeled data for model training.In recent years,some studies have proposed the use of machine reading comprehension models for event extraction,however,the existing methods are limited to single-round question-and-answer mode,which on the one hand ignores the dependency relationship between event theory elements,and on the other hand,does not make full use of the a priori information and other knowledge.In this paper,we propose a question-and-answer mode event extraction method based on the machine reading comprehension framework to address the above problems,construct a multi-round question template and question-and-answer framework,and improve the performance of the model by introducing a priori knowledge and multiround historical answers.The following elements are focused on:(1)In terms of the model framework,an event extraction method based on the question-and-answer model is proposed.The event extraction task is converted into a form based on a series of questions and corresponding answers,and the construction of the trigger word extraction model and thesis element extraction model is carried out under the RoBERTa pre-training model,and the event detection and event element extraction is carried out through the pipeline model,and the event detection is divided into two subtasks of trigger word identification and event classification,and a binary classification model is designed for event type judgment to reduce the cascading error,which and then improve the accuracy of the event detection task.(2)In terms of model input,we construct a multi-round question template and question-answer framework,first introduce a priori knowledge and historical answer information,fill them into the corresponding slots of the multi-round question template,and then use the answer information of the upstream round to constructing subsequent questions,capturing the relationships and hierarchical dependencies between questionanswer pairs.This method can help the model accurately capture contextual information,enable the model to comprehensively understand the questions and accurately infer the correlations among the argument elements,and improve the effectiveness and accuracy of event element extraction.Experimental evaluation of the model on the ACE2005 dataset,including comparative experiments and ablation experiments on two subtasks of trigger word recognition extraction and event argument element recognition extraction,all show that the model proposed in this paper,which incorporates a multi-round question and answer and machine reading comprehension models,can effectively improve the performance of the event extraction task.