Font Size: a A A

Knowledge Graph Embedding With Triple Context And Text

Posted on:2019-02-23Degree:MasterType:Thesis
Country:ChinaCandidate:J ShiFull Text:PDF
GTID:2428330596960875Subject:Software engineering
Abstract/Summary:PDF Full Text Request
As the development of deep learning,a new method named knowledge graph embedding,is attracting more and more attention.Knowledge graph embedding aims to encode the semantics of entities and relations in a continuous vector space.It can efficiently measure semantic correlations of entities and relations,alleviate sparsity issue,and significantly improve the performance of knowledge acquisition,fusion and inference.However,most existing methods handle each triple independently,ignoring rich features in the graph which have been proved to be helpful for inference in knowledge graph.Additionally,there is abundant semantic information of entities and relations contained in corpuses,such as news releases and Wikipedia articles,which are also helpful for modeling triples in a knowledge graph.Based on the motivation above,this thesis proposes a knowledge graph embedding model based on triple context and text.For each triple,two kinds of structural information in the graph are considered as its context: one is neighbor context,which is the outgoing relations and neighboring entities of an entity,the other is path context,which is a set of relation paths between a pair of entities.Both of these two kinds of context provide useful information for inference.In addition,this thesis utilizes textual information as a supplement to semantics in knowledge graphs.Specifically,this thesis uses the sentences containing mentions of head entity and tail entity in a triple as the descriptions of latent relations between the entity pair.The main contributions of this thesis are as follows:1)Proposing a knowledge graph embedding model with triple context,which uses neighbor context of entities and path context of entity pairs to help model triples.2)Proposing a knowledge graph embedding model with triple context and text,which incorporates the text information between the mentions of entity pairs to model the relation.3)Performing experiments on benchmarks and evaluating the models using widely-used criterias,the results show that the proposed method outperforms several state-of-the-art models in a few aspects.
Keywords/Search Tags:Knowledge Graph, Representation Learning, Knowledge Representation, Distributed Representation
PDF Full Text Request
Related items