A knowledge graph representation learning method based on semantic vector

Abstract

A knowledge map representation learning method based on semantic vectors, comprising the following steps: 1) construction of semantic vectors fused with text corpus; 2) construction of semantic vectors fused with text corpus and knowledge map context; 3) construction of semantic matrix, the process is as follows: Taking the semantic vectors of triples and relations as input, the semantic matrix corresponding to each relation is obtained; 4) Modeling and training, the process is as follows: a new scoring function is designed to construct the embedded representation of entities and relations in the knowledge map model to obtain the embedded representation model of the knowledge graph; use the stochastic gradient descent method to train the embedded representation model to minimize the value of the loss function, and obtain the semantic vectors of entities and relationships in the final knowledge graph. The representation learning cube proposed by the present invention can relatively model the complex relationship of the knowledge graph, and can improve the accuracy of vector representation.

Description

technical field [0001] The invention relates to the fields of knowledge graph, representation learning, semantic information, etc., and in particular provides a semantic vector-based knowledge graph representation learning method. Background technique [0002] Knowledge graph representation learning aims to map all entities and relationships to this space and retain their original attributes by constructing a continuous low-dimensional vector representation space, so that a large number of efficient numerical calculation and reasoning methods can be applied to better solve data sparseness And the problem of computational inefficiency is of great significance to knowledge map completion and reasoning. [0003] The translation-based representation learning model TransE (Annual Conference on Neural Information Processing Systems, 2013) is an important representation learning method proposed in recent years. This model regards the relation r as a translation vector from the head...

AI Technical Summary

Problems solved by technology

The TEKE (International Joint Conference on Artificial Intelligence, 2016) model builds a co-occurrence network of words and entities based on text corpus to obtain entity description information. The relationship description information is the intersection of the description information of the head entity and tail entity of the triplet, so the relationship is in There are different representations in different triples, which solves the problem of complex relationship modeling in knowledge graphs, but this method involves too many text words, resulting in too long training time

Images