Skip to content
/ PGPR Public

Reinforcement Knowledge Graph Reasoning for Explainable Recommendation

Notifications You must be signed in to change notification settings

orcax/PGPR

Repository files navigation

Reinforcement Knowledge Graph Reasoning for Explainable Recommendation

This repository contains the source code of the SIGIR 2019 paper "Reinforcement Knowledge Graph Reasoning for Explainable Recommendation" [2].

Datasets

Two Amazon datasets (Amazon_Beauty, Amazon_Cellphones) are available in the "data/" directory and the split is consistent with [1]. All four datasets used in this paper can be downloaded here.

Requirements

  • Python >= 3.6
  • PyTorch = 1.0

How to run the code

  1. Proprocess the data first:
python preprocess.py --dataset <dataset_name>

"<dataset_name>" should be one of "cd", "beauty", "cloth", "cell" (refer to utils.py).

  1. Train knowledge graph embeddings (TransE in this case):
python train_transe_model.py --dataset <dataset_name>
  1. Train RL agent:
python train_agent.py --dataset <dataset_name>
  1. Evaluation
python test_agent.py --dataset <dataset_name> --run_path True --run_eval True

If "run_path" is True, the program will generate paths for recommendation according to the trained policy. If "run_eval" is True, the program will evaluate the recommendation performance based on the resulting paths.

References

[1] Yongfeng Zhang, Qingyao Ai, Xu Chen, W. Bruce Croft. "Joint Representation Learning for Top-N Recommendation with Heterogeneous Information Sources". In Proceedings of CIKM. 2017.

[2] Yikun Xian, Zuohui Fu, S. Muthukrishnan, Gerard de Melo, Yongfeng Zhang. "Reinforcement Knowledge Graph Reasoning for Explainable Recommendation." In Proceedings of SIGIR. 2019.

About

Reinforcement Knowledge Graph Reasoning for Explainable Recommendation

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages