Inductive Knowledge Graph Reasoning for Multi-batch Emerging Entities

This is the official code release of the following paper:

Yuanning Cui, Yuxin Wang, Zequn Sun, Wenqiang Liu, Yiqiao Jiang, Kexin Han, Wei Hu. Inductive Knowledge Graph Reasoning for Multi-batch Emerging Entities. CIKM 2022.

Quick Start

Installation

Install PyTorch following the instructions on the PyTorch. Our code is written in Python3.

Run the following commands to install the required packages:

pip install -r requirements.txt

Dataset

unzip data.zip

It will generate three dataset folders in the ./data directory. In our experiments, the datasets used are: WN-MBE, FB-MBE and NELL-MBE. In each dataset, there are six add_x folders, where add_1 is the validation set of original KG, and add_2~6 are emerging batches.

Following Multi-Hop, the pagerank files are generated using the tool (https://github.com/timothyasp/PageRank).

Update (2022.12.26): Delete the isolated entities in the validation sets. Now, there are 7636, 31406, and 18501 facts in the validation sets of WN-MBE, FB-MBE, and NELL-MBE, respectively.

Training and evaluation

1. Train model

   python src/experiments.py --train --dataset <dataset-name> --gpu <gpu-ID>

   You can use following datasets: WN-MBE, FB-MBE and NELL-MBE.

During training, if you want to see the results on emerging batches, you can use the --run_analysis.

python src/experiments.py --train --run_analysis --dataset <dataset-name> --gpu <gpu-ID>

The model will be evaluated on both validation and emerging batch testing sets during training. Note that due to the different density of entity links, there will be a gap between the results on the validation and test sets.

2. Evaluate model

   python src/experiments.py --inference --dataset <dataset-name> --gpu <gpu-ID>

   You can use following datasets: WN-MBE, FB-MBE and NELL-MBE.

Change the hyperparameters

To change the hyperparameters and other experiment set up, start from the parse_args files and config files.

If you have any difficulty or question in running code and reproducing experimental results, please email to yncui.nju@gmail.com.

Citation

If you find the repository helpful, please cite the following paper

@inproceedings{cui2022inductive,
  title = {Inductive Knowledge Graph Reasoning for Multi-batch Emerging Entities},
  author = {Cui, Yuanning and 
            Wang, Yuxin and 
            Sun, Zequn and 
            Liu, Wenqiang and 
            Jiang, Yiqiao and 
            Han, Kexin and 
            Hu, Wei},
  booktitle = {CIKM},
  year = {2022}
}