Research On Entity And Relation Joint Extraction In Parameter Sharing Model

In daily life,a large amount of information is transmitted by natural language texts.Information extraction is the way that people obtain important information from unstructured texts.Named entity recognition and relation extraction are two basic tasks for the acquisition of entity and relation information.By sharing parameters,people combine the two tasks to pursue higher performance and improve efficiency.Existing joint methods produce redundant information which increases the complexity of calculation and interferes with model training.In addition,named entity recognition is a subtask of this work.The practical application can promote the development of joint extraction.Hereby are the major works towards these contents.An end-to-end neural network model and a tag system are proposed to extract entity and relation jointly.Some existing joint models obtain syntax information by using NLP tools that may lead to error propagation.Neural network can capture language feature to break the limitation.A dynamic memory cell is designed for the task,which can capture and store relation information dynamically for relation detection.At the same time,the stored relation information can be used to help the named entity recognition to boost the performance.Construct a domain-orinted natural language analysis system based on named entity recognition.Comparing the performance of different encoders and decoders to analyze the technology.