Knowledge graph representation learning method fusing graph structure information

Abstract

The invention discloses a knowledge graph representation learning method fusing graph structure information, which comprises the following steps: iteratively calculating triple neighborhood context information by adopting a multi-layer message passing mechanism, obtaining relation path information by adopting a path weight method, and combining an attention mechanism to obtain relation path information; using triple neighborhood context information to measure weights of all paths of the relation, fusing and calculating a final comprehensive path vector to fuse rich graph structure information in a triple vector matrix, training through a convolutional neural network to obtain a missing target entity, and complementing the triple and the knowledge graph; according to the method, more accurate and effective path vector representation is obtained, the influence between the path vector and the neighborhood context is enhanced, and the representation learning ability of the knowledge graph is improved.

Description

technical field [0001] The invention relates to a representation learning method of a knowledge graph. Background technique [0002] The knowledge graph is a multi-relational knowledge base that describes many objective facts in the form of triples (h, r, t). In recent years, knowledge bases such as Freebase, YAGO, and DBpedia have been constructed to support applications such as intelligent question answering, web search, and recommendation systems. Although there are a large number of high-quality triplet facts in the knowledge graph, for example, about 71% of people in Freebase lack birthplace, and about 75% of people have no nationality information, so there is knowledge incompleteness in the knowledge graph, these applications The performance of is also affected by the completeness of the knowledge graph. Knowledge map completion is to use the existing knowledge map information to mine potential objective facts and complete the missing information in the knowledge map...

AI Technical Summary

Problems solved by technology

However, there are deficiencies in the existing path modeling methods. For example, when using RNN and other models to model relationship path information, the length of the relationship path is often kept between 2 and 4 hops. Because the length of the relationship path is not long enough, these depths are used. The learning model does not guarantee that the output path representation is sufficiently effective for the path sequence training, and the simple relationship vector method coarsens the weight of each relationship, which is equivalent to setting the weight to 1 and equal

Therefore, none of the existing knowledge map representation learning methods can obtain effective path vector representations, resulting in insufficient knowledge representation capabilities, thus affecting the completion of knowledge maps

Images