GCKB: A novel Gated Convolutional Embedding Method for Knowledge Graphs

Source code for paper, GCKB: A novel Gated Convolutional Embedding Method for Knowledge Graphs.

Requirements

torch >= 1.6.0

Datasets

We used two commonly benchmark datasets for evaluating GCKB.

FB15k-237,

WN18RR.

Training and Evaluation

Parameters

--data: Specify the folder name of the dataset.

--epochs: Number of epochs.

--lr: Initial learning rate.

--weight_decay_conv: L2 regularization.

--output_folder: Path of output folder for saving models.

--batch_size: Batch size

--valid_invalid_ratio_conv: Ratio of valid to invalid triples for training.

--out_channels: Number of output channels in convolution layer.

Training log

log.txt

The sprcific process can be viewed in the "checkpoints" folder.