Relation Extraction Based On Convolutional Neural Network

Relation extraction is a very important branch of natural language processing and a key technology for constructing knowledge maps.Relationship extraction refers to the automatic recognition of the semantic relationships contained between two entities in a sentence containing two entities.Distant supervision and deep learning was introduced into this task due to the lack of labeled training data.This paper studies the convolutional neural networks based relational extraction technology,focuses on the structural design of convolutional neural networks,and attempts to improve the performance of relational extraction systems with a new network structure.This paper first introduces the background and research significance of relation extraction,then summarizes all the methods of relational extraction tasks,and analyzes the advantages and disadvantages of these algorithms in detail.Based on this,two improvements are proposed:1)Get more useful features by combining the relatively low-level features represented by word vectors and position vectors with features extracted from multilayers convolutional neural networks.2)To improve the traditional shallow text convolutional neural networks,a method of using dense connections to increase the convolutional layers is proposedExperiments were conducted on large public data sets.The experimental results show that the proposed model can improve the precision of relation extraction.At the same time,the proposed feature fusion can greatly shorten the training time of the model.Through multiple comparison experiments,we have proved that the proposed model is superior to other current models such as PCNN+ATT、CNN+ATT.