seungryeol-22/KG-to-KG-Semantic-Similarity-Baselines

KG-to-KG Semantic Similarity Baselines

This repository contains baseline code for the KG-to-KG Semantic Similarity benchmark dataset:

https://huggingface.co/datasets/seungryeol-22/KG-to-KG-Semantic-Similarity

The code preserves the original baseline pipeline:

scoring -> ranking -> metric

The original scripts expect legacy local input folders such as:

cc_news_graph/
wikitext_graph_data/
cc_news_graph_verbalized/
wikitext_graph_data_verbalized/

The Hugging Face dataset stores the same graph pairs as Parquet files. Run the materialization script first to recreate the legacy folders from the dataset repository.

Prepare Data

python scripts/materialize_legacy_dataset.py --output . --force

This downloads seungryeol-22/KG-to-KG-Semantic-Similarity and creates:

cc_news_graph/
wikitext_graph_data/
cc_news_graph_verbalized/
wikitext_graph_data_verbalized/

Run Pipeline

Run a scoring script:

python scoring/base-kernel-scoring.py

Then run ranking:

python evaluation/ranking.py

Then run metrics:

python evaluation/metric.py

Results are written by the original scripts under Result/ and Metric_Results/.

Code Layout

scoring/
evaluation/
scripts/materialize_legacy_dataset.py

Dependencies

Install the packages needed by the baseline you plan to run. The adapter requires:

pip install datasets

Graph-kernel baselines require:

pip install grakel tqdm numpy

SBERT baseline requires:

pip install sentence-transformers tqdm numpy

KGE and InGram baselines require PyTorch-related dependencies such as torch, torch-geometric, scipy, scikit-learn, and igraph.

Paper and Dataset

Paper: https://arxiv.org/abs/2606.29180

Dataset: https://huggingface.co/datasets/seungryeol-22/KG-to-KG-Semantic-Similarity

Code source: https://github.com/SeungRyeolBaek/KG-to-KG-Semantic-Similarity