Research And Implementation Of Chinese Text Entity Relation Extraction Based On Deep Learning

With the rapid development of the Internet,the size of the text corpus has been enormously lengthened,and the information contained in the text is ex-ploding.In order to cope with the challenge of information explosion,entity relation extraction came into being.The traditional methods of supervised re-lation extraction relies on a large amount of manually labeled data,which cost a lot of time and effort.In order to relax this restriction,entity relation extrac-tion based on distant supervision was born.Distant supervision is based on the assumption that if two entities have a relationship in the knowledge base,then all sentences containing the two entities will represent this relationship.Al-though distant supervision is an excellent strategy for automatically generating training data,noise data is inevitably introduced because its strong assumption.In recent years,deep learning technology has become a research hotspot in the field of machine learning,and has achieved significant progress in the fields of computer vision and natural language processing.This paper introduced deep learning technology into the task of extracting Chinese relation extraction based on distannt supervision,and carried out the following research work:1)This paper proposed a novel Hierarchical attention-based Bidirectional Gated recurrent neural network integrated with entity Descriptions(denoted by HBGD)to select valid instances and capture vital semantic information in them.Furthermore,we incorporated entity descriptions extracted from Wikipedia into the hierarchical attention model to provide supplementary background knowl-edge.The proposed architecture can not only combat the noise introduced by distant supervision,but also adequately extract latent and helpful background information.The experimental results on both Chinese and English datasets show that the proposed approach consistently achieves significant improve-ments on relation extraction as compared with strong baselines.2)This paper proposed a Multi-lingual Attention-based Neural network Relation Eextraction algorithm(denoted by MARE).The key motivation of this model is that relational facts usually have some kind of pattern expression in various languages,and the patterns between different languages are different.Models can take advantage of such pattern information in multi-lingual data to improve the performance of relational extraction.The multi-lingual attention mechanism consists of two types of attention networks.One is the double-level attention in a single language to capture the semantics most relevant to the ex-traction task in the sentence and combat the noise introduced by the distant supervision,and the other is cross-lingual attention which used to pay attention to the consistency and complementarity of information in multiple languages,thus effectively enhancing the learning of relational patterns.The experimental results in Chinese and English datasets show that the proposed method is bet-ter than the single language model and can adapt to the cross-lingual relation extraction.3)This paper built a visual Chinese entity relation extraction prototype system based on deep learning technology.The front end of the system allows the user to search for related entities,and the system returns the entity relation triples and visually displays them in the form of graphs.The backend of the system will automatically crawl the Internet text data periodically,extract the relation triples and store them in the Neo4j database.