DeepPath: A Reinforcement Learning Method for Knowledge Graph Reasoning

DeepPath: метод обучения с подкреплением для логического вывода в графах знаний
William Yang Wang, Wenhan Xiong, Thien Hoang
2017-01-01

Freebaseknowledge graph embeddingsknowledge graph reasoningmulti-hop relational pathsreinforcement learning
We study the problem of learning to reason in large scale knowledge graphs (KGs).More specifically, we describe a novel reinforcement learning framework for learning multi-hop relational paths: we use a policy-based agent with continuous states based on knowledge graph embeddings, which reasons in a KG vector space by sampling the most promising relation to extend its path.In contrast to prior work, our approach includes a reward function that takes the accuracy, diversity, and efficiency into consideration.Experimentally, we show that our proposed method outperforms a path-ranking based algorithm and knowledge graph embedding methods on Freebase and Never-Ending Language Learning datasets.1
1
A policy-based agent uses continuous states derived from knowledge graph embeddings to sample promising relations while extending reasoning paths.
2
DeepPath introduces a reinforcement learning framework for learning multi-hop relational paths in large-scale knowledge graphs.
3
Experiments on Freebase and Never-Ending Language Learning datasets show DeepPath outperforming a path-ranking algorithm and knowledge graph embedding methods.
4
The reward function jointly accounts for path accuracy, diversity, and efficiency, unlike prior approaches focused primarily on accuracy.

multi-hop relational paths in large-scale knowledge graphs

learning to reason over these paths, with accuracy, diversity, and efficiency of path selection as the focal aspects

Publication Details
Publication Date
2017-01-01
Journal
Publisher
ISSN
Cited by
859
Access Type
Author Information
Authors
William Yang Wang
Wenhan Xiong
Thien Hoang
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%