Multi-level Recommendation Reasoning over Knowledge Graphs with Reinforcement Learning

  • Xiting Wang ,
  • Kunpeng Liu ,
  • Dongjie Wang ,
  • Le Wu ,
  • Yanjie Fu ,

The Web Conference 2022 |

Related File

Knowledge graphs (KGs) have been widely used to improve recommendation accuracy. The multi-hop paths on KGs also enable recommendation reasoning, which is considered a crystal type of explainability. In this paper, we propose a reinforcement learning framework for multi-level recommendation reasoning over KGs, which leverages both ontology-view and instance-view KGs to model multi-level user interests. This framework ensures convergence to a more satisfying solution by effectively transferring high-level knowledge to lower levels. Based on the framework, we propose a multi-level reasoning path extraction method, which automatically selects between high-level concepts and low-level ones to form reasoning paths that better reveal user interests. Experiments on three datasets demonstrate the effectiveness of our method.