Research On Key Technologies Of Deep Learning Based Chinese Entity Relationship Extraction

With the rapid development of Web 2.0 technology and the growing maturity of artificial intelligence technology,a large amount of text data has accumulated under the Internet environment,which has the characteristics of massive,multi-source,heterogeneous and contains extremely rich semantic knowledge.Effective extraction of it can effectively extract the structural knowledge information with potential value,so as to realize the structure of text information This paper focuses on how to effectively extract high-quality and structured triple information from multi-source heterogeneous text data,which is one of the important research contents in the field of information extraction,and has important academic value and industrial application value.At present,information extraction technology mainly includes two parts,named entity recognition(NER)and relation extraction(RE).The existing research methods can be roughly divided into the following categories: supervised information extraction model,semi supervised information extraction model,unsupervised information extraction model and information extraction model based on remote supervision.This paper focuses on supervised information extraction,which is the key technology of Chinese information extraction based on deep learning.The main research work is as follows:1.This paper proposes an information extraction model based on named entity recognition and relationship classification.Firstly,bi-lstm is used to model the semantic information of entity context in text statement,and then conditional random field is used Field(CRF)optimizes the dependency relationship among the annotation sets of bio(begin intermediate other),and finally extracts the features of different entities by using the convolutional neural networks(CNN)to achieve the accurate classification of the relationship between entities.2.This paper proposes an information extraction model based on probability graph,which is mainly based on the sequence to sequence(seq2seq)translation modelto extract information,that is,first using the context information to predict the subject s in the sentence,then using the predicted s information to directly predict the corresponding object o and relationship P;3.This paper proposes a joint information extraction model based on bidirectional encoder representation from transformers(BERT)model + based on multi head mechanism.The model parallelizes the two subtasks of entity recognition and relationship extraction,and effectively solves the problem of multi relationship with entities through interactive promotion;Experiments on real information extraction dataset show that the Chinese information extraction model based on in-depth learning can effectively improve the extraction performance of entity relation triples,which is much better than the existing model.