Research On Entity Relation Extraction Technology Based On Dilate Gate Convolutional Neural Network

With the development of 5G communication technology and information science and technology,people will encounter a lot of intricate information when browsing the web.However,some of this information is needed by people,and some are spam generated by machines.At this time,an automation is particularly needed.Tools to process these data and sort out useful information that meets the needs.Therefore,in this social context,Information Extraction(IE)technology emerged at the historic moment.Entity relationship extraction is a key task of information extraction,and its research and exploration have become particularly important.In recent years,the development of neural networks has promoted the research of entity relationship extraction.Although great achievements have been made,there are still some problems,such as the lack of context and long-distance relationship capture.The research in this paper will explore key technologies based on these existing problems,mainly including the following aspects:1.This thesis combines the gating mechanism with the dilated convolutional neural network and proposes a SADGC algorithm for relation extraction.This algorithm is mainly divided into two steps: the first step is to extract the main entity;the second step is to extract the guest entity and the corresponding relationship.The extraction of the main entity includes: first pass the preprocessed word mixture vector and position vector into the 15-layer expansion gate model for training,and pass the trained coding sequence into self-attention for training.The result is generated,and the generated result plus the prior features obtained from remote supervision are spliced and passed into the fully connected layer of the convolutional neural network to extract the main entity.2.The extraction of the object entity and the corresponding relationship includes:first randomly selecting a labeled main entity during training,reusing the coding sequence in the first step,and extracting the subsequence in the coding sequence;then passing the subsequence to the 6-layer expansion The door model is trained to obtain the coding vector of the main entity randomly selected,and the coding vector and the relative position vector are spliced to obtain a new vector sequence;then the coding sequence in the first step is passed into self-attention for result generation,Splicing the generated result with the new vector sequence and prior features;finally,the splicing result is passed to the fully connected layer of the convolutional neural network to extract the object entity and the corresponding relationship.3.On the basis of the SADGC algorithm,a SABDGT algorithm is proposed for entity relationship extraction.The convolutional neural network in the SADGC algorithm is replaced by Text CNN for result classification.The 6-layer expansion gate model in the extraction of guest entities and corresponding relations is replaced by a two-way long The short-term memory(Bi LSTM)model is used to encode the main entity.Experiments have shown that the algorithm proposed in this paper has achieved significant results.