Knowledge Graph Reasoning with Reinforcement Learning Agent guided by Multi-relational Graph Neural Networks

Chung Wang Wong, Michael Ng

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Knowledge Graphs, Reinforcement Learning, Graph Neural Networks

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: Reinforcement Learning (RL) has emerged as a highly effec- tive technique in various applications, including Knowledge Graph (KG) Completion. KG Completion involves navigat- ing through an incomplete KG from a source entity to a target entity based on a given query relation. However, existing RL- based approaches only focus on training the agent to move along the graph, seldom take into account the multi-relation connectivity inherent in knowledge graphs. In this paper, we propose a novel approach, Reinforcement learning agent Guided by Multi-relation Graph Neural Network(RGMG). Our approach develop a Multi-relation Graph Attention Net- work (MGAT) which generate high quality KG entity and relation embedding to help agent navigation. Additionally, we develop a Query-aware Action Embedding Enhancement (QAE) module to strength information contained in action embedding. Experiments on various KG reasoning bench- marks demonstrate that RGMG is highly competitive and out- performed current state-of-the-art RL-based methods in dif- ferent dataset.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3566