Neural Methods for Logical Reasoning over Knowledge Graphs

This repo contains the code implementing the models described in Neural Methods for Logical Reasoning over Knowledge Graphs

In this paper, we focus on answering multi-hop logical queries on Knowledge Graphs (KGs). To this end, we have implemented the following models. We also include the original baseline models

Baselines

• BetaE
• Query2box
• GQE

Models

• MLP: Multi-Layer Perceptron
• MLPMixer: Adpated from MLPMixer

Variants

• MLP + Heterogeneous Hyper-Graph Embeddings
• MLP + Attention Mechanism
• MLP + 2 Vector Average

How to use it

You can find some examples on how to execute the code can be found on examples.sh

Data

To evalute the models, we have used standard evaluation datasets (FB15k, FB15k-237, NELL995) as in the BetaE paper. It can be downloaded here.

Citations

If you use this repo, please cite the following paper.

@inproceedings{
    amayuelas2022neural,
    title={Neural Methods for Logical Reasoning over Knowledge Graphs},
    author={Amayuelas, Alfonso and Zhang, Shuai and Rao, Xi Susie and Zhang, Ce},
    booktitle={International Conference on Learning Representations},
    year={2022}
}

Acknowledgements

This code is built on top of previous work from SNAP-Stanford. Check out their repo here