Research And Application Of Entity Recognition And Relation Extraction Based On Deep Learning

With the advent of the big data era,a large amount of data stored in unstructured or semistructured forms such as text,audio,and images have been generated in human-computer interaction.Most of them are text data.In order to effectively extract valuable information in text for people to use,entity recognition and relation extraction technologies are two basic tasks in the field of natural language processing,which are designed to help people dig out names and names that have entity meaning in text.The relationship between them is of great significance and value for the construction of domain knowledge graphs and intelligent question answering systems.The traditional relationship extraction task divides entity recognition and relationship extraction into two independent subtasks,and does not consider the dependency transitivity between the two tasks and the information redundancy of the two encodings of the input data,which affects the efficiency and efficiency of the extraction.Aiming at the problems of the pipelined extraction strategy,this article mainly studies how to extract the two tasks jointly.The main innovations and work are as follows:Considering the establishment of a language model has a direct impact on the mining of semantic information within a sentence,this paper uses the XLNet language model to model the word vector of the input data,but it does not directly use the trained word vector or word vector.It uses two-way LSTM to extract features from the input word vector.When the prediction matrix detects that the features extracted by the forward LSTM and the backward LSTM are the same word,the probability of the same word is the largest,and it is fused into the word vector feature by matrix operation.This vector representation strategy not only retains the characteristics of Chinese words into sentences,but also deepens the feature representation of each word in the word.The model is divided into three modules,namely sharing module,entity recognition module and relation extraction module.The shared layer realizes the common encoding of the input text representation of the two tasks.The entity recognition result and the output of the shared layer are used as the input of the relation extraction module.The relation extraction and entity recognition can use the loss function to update the shared layer parameters together using the backpropagation algorithm to achieve joint extraction truly.In January 1998,the People's Daily,COAE2016 task three and self-built small data sets in the economic and financial fields were compared and analyzed experimentally,which verified the effectiveness of the model proposed in this article on the task of entity relationship extraction.Based on the research theoretical results,the knowledge of the Python language and the front-end framework,an entity relationship extraction system is designed mainly for the economic and financial fields.The system can directly extract entity relationship triples in the text,which provides technical support for constructing domain data sets.