This repository contains the code needed to replicate the experiments of the paper "Type-Less yet Type-Aware Inductive Link Prediction with Pretrained Language Models" (EMNLP 2025).

## Setting up the Environment

This repository is based on the following dependencies:

• Python 3.8.19
• CUDA 12.1
• DGL 2.4.0
• PyTorch 2.3.0

To set-up the environment, execute:

conda create -n tyler python=3.8.19
conda activate tyler

To install the required dependencies, execute:

conda install -c dglteam/label/th23_cu121 dgl
conda install pytorch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 pytorch-cuda=12.1 -c pytorch -c nvidia
pip install -r requirements.txt

To train TylerR, run the command (the example is for FB237-V1, but it applies to any of the datasets):

python train.py -e tyler_fbv1_roberta -d fb237_v1 --pretrained_language_model="FacebookAI/roberta-large" --pretrained_language_model_is_causal=False

## Training Baselines

To train GraIL, run the following command:

python train_grail.py -e grail_fbv1 -d fb237_v1

To train the ontology-enhanced model (Zhou et al. 2023), run the following:

python train_oeilp.py -e oeilp_fbv1 -d fb237_v1

Note: it is heavily recommended to train TyleR before any other baseline.

To run evaluation, run the command (Note: it requires a pre-existing checkpoint, which can be obtained by running the training script*)

python test_ranking.py -e tyler_fbv1_roberta -d fb237_v1_ind

For yago, run the following:

python test_ranking.py -e tyler_yago_roberta -d yago --use_test_subgraph=True

• Komal K. Teru, Etienne G. Denis, and William L. Hamilton. 2020. Inductive relation prediction by subgraph742 reasoning. In ICML, volume 119 of Proceedings of Machine Learning Research, pages 9448–9457. PMLR.

• Wentao Zhou, Jun Zhao, Tao Gui, Qi Zhang, and Xuanjing Huang. 2023. Inductive relation inference of knowledge graph enhanced by ontology information. In EMNLP (Findings), pages 6491–6502. Association for Computational Linguistics