Research And Implementation Of Entity Relation Extraction Based On Deep Learning

The 21st century is undoubtedly the era of big data,and all kinds of data information generated in people’s lives contain enormous value.Most of these information exist in the form of electronic documents,and how to mine effective information from these unstructured data has become a problem worthy of study.The information extraction(Information Extraction)research is a topic to solve this problem.The main task of information extraction is to extract specific event or fact information from natural language text,which includes entity(Entity),relationship(Relation))and events.After obtaining this information,it can be saved in a structured form for the convenience of downstream tasks.Information extraction includes multiple subtasks,and relation extraction is one of the important subtasks.Relation extraction can be divided into sentence-level relation extraction and document-level relation extraction according to the length of the input.Most of the previous researches focus on sentence-level relation extraction.However,in application scenarios,many entity relations can only be obtained by combining information inferences from multiple sentences,so the research on document-level relation extraction is obviously more valuable.The task of document-level relation extraction,which contains more information,also brings a series of new problems,so document-level relation extraction requires more complex reasoning,such as logical reasoning,coreference reasoning and synonym reasoning.A document generally includes multiple entities,and an entity has multiple mentions.To identify relationships between entities appearing in different sentences,document-level relation extraction models must be able to model complex interactions between multiple entities and aggregate contextual information from multiple mentions.This paper conducts in-depth research on document-level relation extraction through the investigation and study of the basic theory and development of relation extraction technology,as well as graph generation methods and deep learning techniques such as convolutional neural networks.The main content of this paper includes the following three aspects:·A relation extraction model for automatically constructing document graph structure is proposed.Compared with the previous rule-based document graph or the document graph based on matrix tree automatic composition,this model does not rely on an external syntax analyzer at all.The edge structure allows the model to explore as many effective edge structures as possible,achieving a competitive effect with a smaller memory usage,and the proposed graph construction method can also be applied in other model which need to construct graph from unstructured document text.·Aiming at the special relation extraction task when one entity is the attribute value of another entity in the input entity pair of relation extraction,a person attribute extraction algorithm based on Attention-Comprehension OpenTag is proposed.Person attribute extraction has the problem of polysemy and difficulty in distinguishing between attribute categories.Based on the Attention-Comprehension OpenTag algorithm,this paper uses the attention mechanism to combine the information of attribute categories to deal with the difficulties in character attribute extraction.·An entity relation extraction system based on the above algorithm is implemented.The system includes three functions,which are document-level relation extraction function,important entities display during documentlevel relation extraction and person attribute extraction.