This is PyTorch & DGL implementation for the paper:

Xiang Wang, Xiangnan He, Yixin Cao, Meng Liu and Tat-Seng Chua (2019). KGAT: Knowledge Graph Attention Network for Recommendation. Paper in ACM DL or Paper in arXiv. In KDD'19, Anchorage, Alaska, USA, August 4-8, 2019.

You can find Tensorflow implementation by the paper authors here.

Knowledge Graph Attention Network (KGAT) is a new recommendation framework tailored to knowledge-aware personalized recommendation. Built upon the graph neural network framework, KGAT explicitly models the high-order relations in collaborative knowledge graph to provide better recommendation with item side information.

If you want to use codes and datasets in your research, please contact the paper authors and cite the following paper as the reference:

@inproceedings{KGAT19,
  author    = {Xiang Wang and
               Xiangnan He and
               Yixin Cao and
               Meng Liu and
               Tat{-}Seng Chua},
  title     = {{KGAT:} Knowledge Graph Attention Network for Recommendation},
  booktitle = {{KDD}},
  pages     = {950--958},
  year      = {2019}
}

The code has been tested running under Python 3.6.8. The required packages are as follows:

• torch == 1.3.1
• dgl-cu90 == 0.4.1
• numpy == 1.15.4
• pandas == 0.23.1
• scipy == 1.1.0
• sklearn == 0.20.0

• FM

python main_nfm.py --model_type fm --dataset amazon-book

• NFM

python main_nfm.py --model_type nfm --dataset amazon-book

• BPRMF (train on multi-GPUs)

python -m torch.distributed.launch main_bprmf.py --dataset amazon-book

• ECFKG (train on multi-GPUs)

python -m torch.distributed.launch main_ecfkg.py --dataset amazon-book

• CKE (train on multi-GPUs)

python -m torch.distributed.launch main_cke.py --dataset amazon-book

• KGAT

python main_kgat.py --dataset amazon-book

With my code, following are the results of each model when training with dataset amazon-book.

Final results on all test users

• FM

  • Proposed in Fast context-aware recommendations with factorization machines, SIGIR2011.

• NFM

  • Proposed in Neural Factorization Machines for Sparse Predictive Analytics, SIGIR2017.

• BPRMF

  • Proposed in BPR: Bayesian Personalized Ranking from Implicit Feedback, UAI2009.
  • Key point:
    • Replace point-wise with pair-wise.

• ECFKG

  • Proposed in Learning Heterogeneous Knowledge Base Embeddings for Explainable Recommendation, Algorithm2018.
  • Implementation by the paper authors: https://github.com/evison/KBE4ExplainableRecommendation
  • Key point:
    • Introduce Knowledge Graph to Collaborative Filtering

• CKE

  • Proposed in Collaborative Knowledge Base Embedding for Recommender Systems, KDD2016.
  • Key point:
    • Leveraging structural content, textual content and visual content from the knowledge base.
    • Use TransR which is an approach for heterogeneous network, to represent entities and relations in distinct semantic space bridged by relation-specific matrices.
    • Performing knowledge base embedding and collaborative filtering jointly.

• KGAT

  • Proposed in KGAT: Knowledge Graph Attention Network for Recommendation, KDD2019.
  • Implementation by the paper authors: https://github.com/xiangwang1223/knowledge_graph_attention_network
  • Key point:
    • Model the high-order relations in collaborative knowledge graph to provide better recommendation with item side information.
    • Train KG part and CF part in turns.