Research On Chinese Personal Relation Extraction Based On BiGRU

Information extraction technology can extract valuable information from unstructured data,and its results can be used in many fields such as knowledge question answering and knowledge graph construction.As an important part of information extraction technology,named entity recognition and personal relation extraction have important research significance and application prospect.In this thesis,we improve a named entity recognition model based on Bi GRU,which is used in the construction of relation extraction data set.Moreover,we add the attention layer of relation words to improve the Bi GRU-ATT model in the processing of personal relation extraction.The main research contents are as follows.Firstly,we improved named entity recognition model based on Bi GRU.To solve the problem that Recurrent Neural Network(RNN)cannot obtain the long distance context information of text,we use ALbert-Bi GRU-CRF model for Chinese named entity recognition,which can not only solve the long distance dependency of RNN model,but also obtain the context information of text.Experiments show that the accuracy rate and recall rate of our model are comparable to current popular models,but the training speed is greatly improved under different EPOCH and BATCH?SIZE.Secondly,we extract and preprocess the personal relation data set.Aiming at the problem of lack of corpus for Chinese personal relation,we first extract nearly 200000 personal relation from Baidu encyclopedia,and then summary their categories and integrate the data set.Finally,ALbert-Bi GRU-CRF was used to eliminate the text sentences of multiple people in the data set.As a result,a high quality data set is obtained.Finally,we improved the personal relation extraction model based on Bi GRU-ATT model.According to the characteristics that the relation word in a sentence can represent the relationship between two people,the relation dictionary is constructed and the weight of relation words is calculated,so that our model can assign higher weight to the text containing the relation words and less weight to the text without the relation words,and focus on learning semantics of the sentence with higher weight.Finally,the Bi GRU layer extracts the context information features of sentences and further enhances their features with the attention mechanism.Compared with Bi GRU-ATT model,the F1 value of our model is improved by 9%.