KG-Hi-BKF Benchmark for Biomedical Knowledge Fusion

The KG-Hi-BKF benchmark datasets for the biomedical knowledge integration task, which is proposed in our SIGIR'2023 paper: "HiPrompt: Few-Shot Biomedical Knowledge Fusion via Hierarchy-Oriented Prompting". There are two datasets: SDKG-DzHi repoDB-DzHi Under each dataset, the files are organized as follows: ent_attr_KG.json ent_attr_Hi.json rel_trip_KG.csv hypo_hyper_pair_Hi.csv ent_links.csv zero_shot/ |-- test_ent_links.csv one_shot/ |-- support_ent_links.csv |-- test_ent_links.csv where `ent_attr*.json` denotes entity attribute files, `rel_trip_KG.csv` denotes knowledge graph relational triples, `hypo_hyper_pair_Hi.csv` denotes hierarchy hyponym-hypernym term pairs, and `ent_links.csv` denotes all aligned entity-term pairs. The two subfolders `zero_shot/` and `one_shot/` provide the support&test split used in the paper. If you use the datasets, please cite our paper ```BibTex @inproceedings{lu23HiPrompt, title = {HiPrompt: Few-Shot Biomedical Knowledge Fusion via Hierarchy-Oriented Prompting}, author = {Lu, Jiaying and Shen, Jiaming and Xiong, Bo and Ma, Wengjing and Staab Steffen and Yang, Carl}, year = {2023}, month = {Apr.}, Series = {SIGIR 2023}, Booktitle = {46th International ACM SIGIR Conference on Research and Development in Information Retrieval - Short Paper}, } ```