Research On Fine-grained Entity Classification Method Based On Deep Neural Network

In recent years,fine-grained entity classification has become an important sub-task in the field of named entity recognition.However,due to the uneven quality of data sets and the absence of a unified fine-grained entity classification type standard,the existing algorithms can not fully consider the hierarchical structure of fine-grained entity types,the semantic information of entity types themselves,and the recognition errors of long entity types in text.Therefore,this paper innovates the deep neural network to improve the performance of fine-grained entity classification.The main research work and results of this paper are summarized as follows.(1)The existing research on fine-grained entity classification method focuses on better encoding the semantic information of reference and context,while ignoring fine-grained entity types have a hierarchical structure.Therefore,this paper proposes a new Hierarchy-Aware Fine-Grained Entity Typing(HAFGET)method.Firstly,the hierarchical encoder based on graph convolution network was used to encode the dependency between different levels of labels.Then the hierarchical structure perception and classification of the context features were carried out by using multi-label attention model and mention feature propagation model.In the former,the hierarchical perceptual label embedded representation was learned through the hierarchical encoder and to learn classified after attention fusion with the entity features.In the latter,the entity features were directly inputted into the hierarchical encoder to update the feature representation and then classified.Experiment results on FIGER,Onto Notes and KNET datasets show that the Accuracy and Macro F1 values of proposed model are both improved by more than 2% compared with the baseline model.It is verified that the proposed model can effectively improve the classification effect.(2)In fine-grained entity classification tasks,the existing methods are not effective for long entity classification,and the semantic information of entity context plays an important role in entity classification.Therefore,in order to better capture the semantic information of long entities,this paper proposes a neural network based on attention convolution to extract the semantic features of long entities.This method makes full use of the advantages of convolutional neural network and attention mechanism,and can not only capture high-level semantic information of entities,but also extract temporal information of sentences.N-gram features of text are extracted by appropriate convolution kernel,and then different weights are assigned to context by attention mechanism,and finally the final representation of entity is generated.The performance of the proposed model was evaluated on Onto Notes and FIGER datasets.Experimental results show that Macro F1 values are improved by3.75% and 1.23%,which verifies the validity of the proposed model.(3)Finally,from the practical application,the development of fine-grained entity classification prototype system is realized,and the fine-grained entity classification model is encapsulated.By using B/S architecture mode for system design,and using Lay UI +Spring Boot+Django+REST Framework technology for system development,so as to visualize the results of fine-grained entity classification task.