Research On Event Extraction Algorithm Based On Sequence Labeling Model

Event extraction is one of the important tasks in the field of information extraction.Its main task is to extract structured event information from unstructured information.At present,with the rapid development of the Internet,Internet text data has also developed spurt.So the extraction of structured event information has extremely important significance.The event extraction task is divided into two subtasks:event type detection and event argument extraction.Traditional methods take them as classification tasks to solve,which are mainly divided into two ways:pipeline and j oint,relying on manually extracted features or deep learning to complete tasks.This research topic is based on the deep learning algorithm,which converts these two subtasks into sequence labeling tasks by pipeline method.The main research contents and results are as follows:1)In the event type detection task,a sequence labeling model based on prior knowledge and Self-attention mechanism is proposed.Use prior knowledge to control the learning process of the end-to-end model and improve the interpretability of the model.The trigger word prior knowledge helps the model to reduce the noise caused by non-trigger words and increase the weight of candidate trigger words.The Self-attention mechanism is used to fully understand the interdependence between words within a sentence,and the problem of not fully utilizing event argument information in the process of identifying event trigger in pipeline model is solved,and the accuracy of event type detection is improved.2)In the event argument extraction task,a sequence annotation model based on Self-attention mechanism is proposed.Based on the detected event trigger words and corresponding trigger words,the Self-attention mechanism is used to fully learn the correlation between event argument and trigger words,effectively complete the event argument extraction task.3)Extend the event detection task to the financial sector.Using the method of distant supervision,a Chinese financial public opinion event data set is generated.Then,the sentence-level attention noise reduction mechanism is added to the event type detection model proposed in this research topic to help the model learn more effective labeling data information,and complete the type of public opinion event detection in the financial field.