Research And Application Of Chinese Relation Extraction Algorithm For Constructing Knowledge Graph Of Economic Field

With the rapid development of the information age,the Internet technology has developed explosively in recent years.Meanwhile,the Internet is widely used in all walks of life,and the text data on the Internet is growing explosively.The amount and variety of data available on the Internet makes it easier for people to access knowledge,but the large amount of useless data also makes it more difficult for people to acquire knowledge efficiently and accurately.There is an urgent desire for a technology that can extract useful knowledge from vast amounts of data,so that when people need a certain knowledge,they can get it directly and accurately,without the need for human screening.Knowledge map is born in this case,knowledge map is covered in huge amounts of data on the network structure of knowledge,it is to extract useful knowledge from huge amounts of data,present to the users in the form of entity relation triples structured,let people can rapidly and accurately to acquiring knowledge,whereas entity relation extraction is the core of knowledge map construction technology.Therefore,knowledge map and entity relationship extraction has been a hot research direction since it was proposed,it is of great value and significance in both academic and engineering to study entity relation extraction algorithm for domain knowledge mapping.Firstly,this thesis introduces the development and research status of knowledge mapping and entity relation extraction algorithms at home and abroad.Then this thesis analyzes the related techniques of knowledge mapping and entity relation extraction.Next,on this basis,this thesis makes an in-depth study of entity relationship extraction algorithm based on knowledge map of economic domain,and proposes an improved entity relationship extraction algorithm based on remote supervision.Finally,the improved algorithm model is applied to the actual data in the economic field,and the knowledge map center system of big data in the economic field is successfully built.On the whole,the specific work and contributions of this thesis are as follows:(1)Aiming at the problem of low accuracy and recall rate of the data automatic labeling module of the remote supervised entity relation extraction algorithm,this thesis proposes the automatic labeling algorithm that joints relation feature words and sentences similarity in the data.Firstly,the dependency syntax analysis is used to optimize the sentence similarity,so that the sentence similarity mainly depends on the sentence elements related to the entity pair and emphasizes the constraints of the entity pair on the sentence.Then,the data automatic annotation algorithm in this thesis is realized by combining the improved method of sentence similarity calculation and the traditional method of relational feature word matching.Finally,the improved data autotagging algorithm in this thesis is compared with the classical Mintz method,relational feature word extension method and keyword similarity method through experiments,also the experimental results prove that the data automatic labeling algorithm in this thesis has a great improvement in accuracy compared with the Mintz method and a better improvement in accuracy and recall rate compared with the relational feature word extension method and the keyword similarity method.(2)In view of the existing remote supervision relation extraction model,the relation in the sentence that is constrained by entity pair is ignored,and the classical algorithm PCNN-ATT model for specific Chinese data in the economic field also has some small problems such as Chinese word segmentation noise transfer,multi-relation classification and double calculation.On the basis of the pcnns-att model,this thesis proposes the PCNNs-ATT-DP model by introducing dependency syntax analysis,word mixing vector,multi-label problem and relation representation to realize remote supervised relation extraction.Finally,the improved PCNNs-ATT-DP model in this thesis is compared with the classical Mintz model,MIML model,PCNNs model and PCNNs-ATT model through experiments,and the experimental results proved that PCNNs-ATT-DP model in this thesis,compared with other classic Mintz model,MIML model,PCNNs model and PCNNs ATT model,has a good promotion in terms of accuracy and recall rate for the entity relationship extraction task of Chinese data in the economic field and the highest F value in all models;(3)On the basis of the improved automatic labeling algorithm for economic data in chapter 3 and the improved remote supervision relation extraction model in chapter 4,this thesis also designs and realizes the big data knowledge map center system for specific data in the Chinese economic field,and realizes the application of the research algorithm in this thesis.