Relation Extraction Technology For Enterprise Knowledge Graph Construction

Entity relationship extraction is one of the important tasks in the field of natural language processing,and it is also a key step in the process of building knowledge graphs.The purpose of entity relationship extraction is to extract the relationship between entities from natural language texts,so as to link entities together to form a mesh knowledge base with triples as the knowledge unit,which becomes the knowledge source of knowledge map.In the field of entity relationship extraction,there have been many methods,but there are still various problems.In this paper,through in-depth study of various relationship extraction methods,this paper aims to construct the actual goal of enterprise map,and proposes two relationship extraction models based on deep learning.Because the current research is mainly based on English data sets,and the Chinese standard data set is lacking.This paper builds a data set for enterprise relationship extraction by means of web crawling and using remote monitoring.The main sources of data are public company announcements and corporate news.In order to make the experiment more convincing,in the final experimental stage,a standard English relation extraction data set was also used.Traditional relation extraction methods usually need to formulate a large number of rules or construct complex feature engineering.With the development of deep learning,the deep learning-based relationship extraction algorithm is gradually proposed.Based on the current research,this paper proposes two improved deep learning-based relationship extraction models.The first model is the BiGRU-CNN model,which combines the BiGRU network and the CNN network in tandem,taking advantage of the CNN network and the RNN network.Another model is the BiGRU-Incep model,which combines the BiGRU network with the Attention mechanism and uses a one-dimensional Inception structure in parallel.For the multi-instance problem of the same entity pair,the method of adding the sentence-level attention mechanism in the relationship extraction is adopted,which reduces the noise influence brought by the remote supervision algorithm and improves the accuracy of the entity relationship extraction.In feature selection,this paper uses pre-trained word vectors as the main feature input,and also uses features such as part of speech and location that are easy to obtain,avoiding complex feature engineering.Experiments have shown that using pre-trained word vectors and adding part-of-speech positions can improve the model's effect.Finally,this paper uses the newly proposed relationship extraction model,combined with other steps of knowledge graph construction,to construct a small enterprise knowledge graph,which has certain practical value.