Research And Implementation Of Chinese Entity Relation Extraction Based On Deep Neural Network

With the rapid development of information technology,the amount of data in the network is increasing day by day.These data usually contain rich and effective information for people to use.In order to fully mine the knowledge contained in it,information extraction tasks are generated in time.Through information extraction,people can easily obtain structured,easy-to-understand,and directly usable data from the original data.Entity relationship extraction is one of the most critical subtasks of information extraction.Traditional entity relationship extraction research deals with this task in a pipelined manner,first identifying entities and then detecting relationships between entities.These methods ignore the correlation between entities and relationships and lead to error propagation.In order to solve this problem,this paper implements a joint extraction method of entities and relationships,which encodes sentences and embeds them in a neural network model.It performs entity recognition and relationship extraction subtasks at a time to generate knowledge that can be used directly,namely,entity-relation triples.In addition,in order to overcome the disadvantage that supervised learning relies on a large number of manually labeled data,the method of distant supervision is introduced to align the knowledge base with unstructured texts to automatically build a large number of training data.Neural networks have a strong ability of expression,which have attracted the attention of researchers in many application fields,and have achieved good results in natural language processing tasks.In this paper,deep neural network is used to modeling the joint task of entity recognition and relationship extraction,and carries out the following research:1)A novel Hierarchical attention-based end-to-end Bidirectional long short-term memory network integrated with Language Modeling objective(denoted by HBLM)is proposed.Different from traditional pipeline methods,the joint learning method performs two subtasks at the same time,which can make better use of the interaction between them without generating redundant information.The language model with auxiliary training objectives learns how to predict the surrounding words of each word in the data set,and encourages the framework to learn more semantic synthesis features without additional training data.In addition,we integrate the hierarchical attention mechanism into the joint extraction model to capture vital semantic information from available texts.A number of comparative experiments on open standard datasets show that the HBLM model is significantly better than the existing mainstream models.2)An Open-Type joint entity and relation extraction via Hierarchical Reinforcement Learning(denoted by OHRL)is proposed.The model integrates two open extractors to extract triples to build knowledge base,and introduces the method of distant supervision to align the knowledge base with unstructured texts,so as to automatically build a large number of datasets.At the same time,the model introduces reinforcement learning to deal with the extended dataset based on distant supervision.For the same sentence,the information is extracted through two levels of reinforcement learning,high-level RL extraction relationship and low-level RL identification entity.Experiments on open standard datasets show that OHRL model has achieved remarkable results and can adapt to the extended datasets based on distant supervision.3)A visualization system of Chinese entity relationship extraction was constructed.The system mainly focuses on the mining of character relationships.The user searches for related information of characters through the front-end interface,then the back-end of the system queries the entity relationship triples from the neo4j database,and visualizes them in the form of figure relationship network.The system automatically crawls the textual data from the Internet on a regular basis,mines the entity relationship information from it,and stores it in the database.