Knowledge graph inference model, system and inference method for Bayesian small sample learning

Abstract

The invention relates to a knowledge graph inference model, system and inference method for Bayesian small sample learning. The method at least comprises the following steps: performing Gaussian distribution modeling on entities and relationships in a knowledge graph to reduce the uncertainty of the knowledge graph; taking each entity as a task to simulate a meta-training process of a newly appearing entity in the dynamic knowledge graph to perform task sampling; constructing a meta-learner based on the graph neural network and performing random reasoning; the meta-learner is trained to represent a newly occurring entity using a support set. According to the invention, the trained model has rapid adaptation capability, and new facts or appearing entities can be deduced without re-training.

Description

technical field [0001] The invention relates to the technical field of knowledge graph reasoning models, in particular to a knowledge graph reasoning model, system and reasoning method for Bayesian small sample learning. Background technique [0002] Large-scale knowledge graphs, such as YAGO, NELL, Wikidata, contain a large amount of factual knowledge. As the background of many applications, with the rapid development of knowledge graph reasoning, the wide application of knowledge graph reasoning in recommender systems, question answering, etc. has also attracted more and more attention. Knowledge graph inference aims to infer new facts from existing knowledge graphs, which are often modeled as a link prediction problem to predict a new entity or relation for a query triple. [0003] There are many different studies that have made substantial contributions to knowledge graph reasoning, and one of the most popular methods is based on knowledge graph embedding methods, which...

AI Technical Summary

Problems solved by technology

The embodiment of this patent performs knowledge reasoning on the initial knowledge map by invoking the knowledge map reasoning model obtained based on reinforcement learning training, without considering the change of time, and uses multiple trainings to improve the knowledge map

Images