Research And Implementation Of News Event Extraction In Public Health Field

Event extraction is an important task in the process of information extraction.As a key technology for constructing event knowledge graphs,it has great application value in public opinion monitoring,information retrieval,intelligent question answering,etc.At present,there are still many challenges about news event extraction in the field of public health.Firstly,there are no publicly available annotated corpora in the field of public health;secondly,the problems of multiple event trigger words and long trigger words make it difficult to extract event trigger words;thirdly,the existing event arguments are separated from the event trigger word extraction task,and the distribution of event arguments is unbalanced.In response to the above problems,this thesis proposes a novel event extraction method in the public health field,which mainly includes the following two aspects:(1)For event trigger extraction subtask,aiming at the difficulty of extracting multi event trigger and long event trigger in public health news and the lack of publicly available datasets,this thesis proposes a novel algorithm about event trigger word extraction based on dependency parsing(Event Trigger Word Extraction Algorithm based on Dependency Parsing,DPETE).First,the algorithm uses related technologies to crawl news corpus from news websites,and obtains a dataset in the public health field through data processing;then it establishes a trigger word list for public health news;finally,it uses the language technology platform of Harbin to perform dependency parsing on sentences and formulates rules to build event-trigger extraction algorithms.The proposed algorithm has been compared with other models on the public health news dataset and the ACE 2005 Chinese dataset,which proves that it is far more effective than any other models.(2)For event argument extraction subtask,aiming at the problem that the existing event argument and event trigger extraction tasks are separated and the distribution of event arguments is unbalanced,this thesis proposes an event argument extraction model incorporating the features of event trigger words(Event Argument Extraction Model with Trigger Features,TFEAE).First,it adds the trigger word distance feature when inputting the model and integrates the trigger word semantic feature through the Conditional Layer Normalization.Then it uses the Chinese Ro BERTa model with better effect to replace the traditional BERT model;finally,the event arguments are grouped and extracted.In relative terms,this method can reduce the impact of the unbalance of event argument distribution.Experiments on the public health news dataset and the ACE 2005 Chinese dataset show that the model can effectively extract the event arguments of the news in the public health field.