Paper Type
Complete
Paper Number
1206
Description
Biomedical knowledge graphs (KGs) play a crucial role in biomedical research and clinical settings. However, large biomedical KGs often contain erroneous statements about biomedical concept due to automated extraction from biomedical literature. These inaccuracies using in downstream applications can compromise the validity of biomedical research studies or lead to erroneous conclusions. This study aims to design an effective method for determining the correctness of triplets in a biomedical KG. We propose a knowledge graph verification (KGV) method, which includes a knowledge graph embedding (KGE) training stage and a KG triplet classification model training stage. We design three different modes for KG triplet classification: Independent, Shared, and Multitask Learning (MTL) mode. Using SemMedDB as the source KG to train a KGE model and a dataset of 3,760 expert-annotated triplets, we empirically evaluate the effectiveness of our proposed KGV method. Our results show that the MTL mode generally outperforms the others.
Recommended Citation
Tsai, Pei Yuan; Wei, Chih-Ping; and Li, Jih Jane, "A Novel Neural Network Architecture for Biomedical Knowledge Graph Verification" (2024). PACIS 2024 Proceedings. 20.
https://aisel.aisnet.org/pacis2024/track11_healthit/track11_healthit/20
A Novel Neural Network Architecture for Biomedical Knowledge Graph Verification
Biomedical knowledge graphs (KGs) play a crucial role in biomedical research and clinical settings. However, large biomedical KGs often contain erroneous statements about biomedical concept due to automated extraction from biomedical literature. These inaccuracies using in downstream applications can compromise the validity of biomedical research studies or lead to erroneous conclusions. This study aims to design an effective method for determining the correctness of triplets in a biomedical KG. We propose a knowledge graph verification (KGV) method, which includes a knowledge graph embedding (KGE) training stage and a KG triplet classification model training stage. We design three different modes for KG triplet classification: Independent, Shared, and Multitask Learning (MTL) mode. Using SemMedDB as the source KG to train a KGE model and a dataset of 3,760 expert-annotated triplets, we empirically evaluate the effectiveness of our proposed KGV method. Our results show that the MTL mode generally outperforms the others.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
Healthcare