Font Size: a A A

Research On Hidden Relationship Discovery Algorithm Based On Deep Learning

Posted on:2021-05-16Degree:MasterType:Thesis
Country:ChinaCandidate:M J ZhaoFull Text:PDF
GTID:2428330632453274Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
With the rapid developments of Internet technologies and popularization of Internet among daily activities,different formats of data continue to emerge,including text,images,video,and audio data,of which text type data dominates in quantity.Hence,to mine valuable information from massive data and knowledge has always been a research hotspot in the field of artificial intelligence.The proposed knowledge graph provides a knowledge model method for information mining.However,due to the incomplete data sources and the omission of knowledge acquisition during the construction of the knowledge graph,the existing knowledge graphs are mostly incomplete.In this environment,the task of constructing and completing knowledge graphs is an important research direction in the field of information mining.In order to fully excavate the existing knowledge graph and the information in the text,the thesis first proposed the concept of hidden relation.Meanwhile,with the continuous developments and maturity of deep learning technology,extracting hidden relationships in the field of natural language processing with deep learning method is been widely used and achieved excellent results.Therefore,the paper mainly studies the hidden relationship discovery algorithm based on deep learning.The thesis mainly includes the following two aspects of work:(1)Discover hidden relationships in existing knowledge graphs.For the existing knowledge graphs,we introduce external entity description information,with a full attention mechanism,and then we effectively integrate structural representation and text representation of the entity,with modification of the model's loss function to improve the representation of entity and relationship.Based on the ideas from translation,we can mine the implicit relationships in the knowledge graph.The model is verified by two classical tasks of constructing an complete knowledge graph by entity prediction and relationship.Compared with the original model,our model demonstrates good experimental results.(2)Extract implicit relations from the text.The paper proposes a new method of joint extraction,which not only avoids the accumulation of errors in the traditional pipeline method,but also takes into account the fact that there will be entity duplication or multiple relationships between entities in the text.This new method used Sequence labeling method.Overall,the effect of the model in the two sub-tasks of entity recognition and relationship recognition was verified through experiments.On the NYT dataset,our model achieved the accuracy,the recall and the F1 score of 80.45%,77.09%and 78.74%,respectively in relationship recognition task.
Keywords/Search Tags:hidden relations, deep learning, knowledge graph, representation learning, relation extraction
PDF Full Text Request
Related items