Research On Relation Extraction Based On Gate Graph Neural Network And Attention

Nowadays,there is a huge amount of text information in structured or semi-structured form,which is of great value.Therefore,it is a very important task to extract valuable and meaningful knowledge from the mass of text information.Information extraction task is born with this goal.This paper studies one of the subtask of information extraction: relation extraction.Relation extraction,which aims at automatically detecting and identifying semantic relations between entities from text,is a key link in the construction of knowledge graph and information extraction,and has important theoretical significance and broad application prospect.The main work of this paper is as follows:1.The Gate Graph Neural Network is introduced into the task of relation extraction,and the semantic features of sentences are extracted effectively with the help of the graph structure of dependency parse tree.In the dependency tree resolution,a clipping algorithm centered on shortest dependency path is used to remove the noise words irrelevant to the relation extraction,and then a gated graph neural network is used to extract the semantic features on the dependency parse tree after clipping.2.A tree distance attention mechanism based on distance in the dependency parse tree is proposed,The weights of different words are calculated by using the distance between the head and tail of the node in the dependency parse tree and the output of the Bidirectional Long Short-Term Memory.The attention mechanism can make use of the information of the dependency parse tree,and can give higher weight to the important words for the relation,which can not only retain the effective information,but also effectively reduce the influence of the noisy words.The paper conducted experiments on the TACRED dataset and the Sem Eval-2010 Task-8 dataset,Compared with the baseline method,the experimental results of the two datasets are good.In addition,ablation experiments are carried out to verify the effectiveness of the innovation work.