Uncertain knowledge graph prediction method based on improved embedded model SUKE

Abstract

The invention relates to an uncertain knowledge graph prediction method based on an improved embedded model SUKE. An SUKE model is provided on the basis of an existing deterministic embedding model DistMult. The SUKE reserves structural information and uncertainty information of knowledge and internally comprises an evaluator and a confidence degree generator, the evaluator evaluates the reasonability of facts according to structural features and uncertainty features of the facts, unreasonable facts are screened out, and therefore candidate facts are obtained. And the confidence coefficient isgenerated for the candidate facts and represents the probability of the specific relationship of the entities. An evaluator defines a structure score and an uncertainty score for each triad for a fact rationality evaluation task. In addition, unknown facts are introduced into the evaluator to participate in training. A confidence generator generates a confidence for each triad for a confidence prediction task. According to the invention, the link prediction task of the uncertain knowledge graph can be effectively completed.

Description

technical field [0001] The invention relates to the technical field of knowledge representation and reasoning under knowledge graphs, in particular to an uncertain knowledge graph prediction method based on an improved embedded model SUKE. Background technique [0002] The uncertain knowledge graph will provide a confidence score for each triple, and the confidence reflects the probability of occurrence of the triple. In recent years, the development of relation extraction and crowdsourcing has promoted the construction of large-scale uncertain knowledge graphs, such as: ConceptNet, Probase, and NELL, etc. Hu et al. proposed URGE in 2017. URGE proposes a matrix factorization based method to embed uncertain networks. But this model only considers node proximity in sparse networks, and only learns node embeddings. UKGE, proposed by Chen et al. in 2019, uses the confidence of triples to learn embeddings, while introducing probabilistic soft logic to infer unseen facts. But ...

AI Technical Summary

Problems solved by technology

For the design of uncertain networks, the URGE model considers the proximity of nodes to generate node embeddings. Although the URGE model can be extended to knowledge graphs, there are differences between uncertain networks and knowledge graphs, and it cannot complete uncertain knowledge graphs well. embedded task

Compared with URGE, UKGE has better performance, but the UKGE model does not make full use of the structural information of knowledge to a certain extent

Images