Knowledge reasoning method and device based on graph representation learning and deep reinforcement learning

Abstract

The invention provides a knowledge reasoning method and device based on graph representation learning and deep reinforcement learning. The method comprises the following steps: constructing a relational graph neural network model, inputting knowledge graph data into the model, and extracting graph topological structure information and semantic information of knowledge according to different relation categories of the input data; and on the basis of the extracted information, constructing a reinforcement learning model, performing knowledge reasoning through interaction of a reinforcement learning agent and an environment, and outputting a reasoning result. Knowledge vectors obtained after graph representation learning contain rich graph topology information and semantic information mainly based on relation categories, powerful single-step reasoning information is provided, and in the reinforcement learning reasoning process, multi-step reasoning is carried out through continuous interaction of an intelligent agent and an environment, so that the reasoning method based on graph representation learning and reinforcement learning can improve reasoning efficiency and enhance reasoning interpretability through complementary combination of single-step reasoning and multi-step reasoning.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, and in particular relates to a knowledge reasoning method and device based on graph representation learning and deep reinforcement learning. Background technique [0002] In recent years, with the rapid development of technologies such as cloud computing and the Internet of Things, the scale of data has shown explosive growth. How to organize and utilize knowledge in data has attracted much attention, and knowledge graphs have emerged as the times require. Today, knowledge graphs have been widely used in search engines, question answering systems, and recommendation systems. At this stage, a large number of knowledge graphs have emerged, and the representative general knowledge graphs include Freebase, DBpedia, NELL, etc. However, due to the openness of the knowledge graph itself and the diversity of construction methods, there are a large number of missing entities and relations...

AI Technical Summary

Problems solved by technology

[0003] The above-mentioned existing knowledge reasoning methods mainly have the following problems: First, the information of the knowledge graph itself has not been fully exploited and utilized, such as graph topology information, attribute information, edge description information, etc. of the knowledge graph; second, with a large number of new methods (such as The introduction of neural network, generative confrontation imitation learning, etc.) increases the model parameters, increases the difficulty of training, and greatly weakens the interpretability of reasoning; third, the robustness of the reasoning method becomes worse, and the improvement of the model tends to solve a certain problem. Specific application problems, or tend to a specific data set, poor model transferability

Images