Research On Chinese Entity Relation Extraction Based On Deep Learning

With the rapid development of the Internet and the popularization of digital devices,the number of Internet users has increased dramatically,and the demand for knowledge services has grown rapidly.However,traditional search engines cannot meet the requirements of accurate knowledge services.Entity relation extraction is a technique to determine the semantic relation between entity pairs in text,and it is the basis of natural language applications such as intelligent question answering system and semantic search.Deep learning has powerful learning ability and has been successfully applied in natural language processing.This thesis aims to study the approaches for Chinese entity relationship extraction based on deep learning.The main work includes the following aspects:1.The technology of web pages crawling in the Chinese Encyclopedia website is studied.Web pages are crawled through Python,XPath and rules are used to extract text data.Technologies for unsupervised word vector training and Distant Supervision data annotating are studied.Word preprocessing is performed on the crawled text data.Then Word2 Vec tools are used for training Chinese word embeddings.The Wikidata knowledge base is automatically aligned with preprocessed texts.Finally,data sets are generated for subsequent research.2.An entity-attention-based deep learning model for entity relation extraction is proposed.This model employs the semantic information of entities during distinguishing the semantic relation between the entity pairs.Firstly,the bidirectional LSTMs are used to model the entity pairs' context information.Secondly,the entity attention model is used to assign different weights to the semantic features that plays different roles in the semantic relations.Then,the computational results are normalized by Softmax as the probabilities of all the relations.Finally,the parameters are optimized by gradient descent method.The experimental results show that the proposed model may improve the effect of entity relation extraction.3.A multi-instance multi-label based deep learning model for entity relation extraction is proposed.It can realize to extract entity relation on the data set with noise.It combines entity relation with multi-instance multi-label learning framework.Firstly,the representation of each sentence is obtained by entity attention mechanism.Then,the noise sentences in the sentences set are filtered through the sentence-level attention.Finally,it extracts the relations between entity pairs.Experimental results indicate that the model is effective for entity relation extraction.