TY - GEN
T1 - Conditional Knowledge Graph
T2 - 8th China Conference on Knowledge Graph and Semantic Computing, CCKS 2023
AU - Lv, Yaojia
AU - Zheng, Zihao
AU - Liu, Ming
AU - Qin, Bing
N1 - Publisher Copyright:
© 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
PY - 2023
Y1 - 2023
N2 - Facts are conditionally established in most cases. However, current Knowledge Graph (KG) techniques only focus on the modeling and representations of facts, neglecting the presence of conditions, which are necessary to establish the validity of facts. In this paper, we propose Conditional Knowledge Graph (Conditional-KG), which employs a three-layer hierarchical network to incorporate both facts and conditions. To facilitate research on the automatic construction of Conditional-KG, we manually annotate an innovative large-scale dataset named HACISU. Based on the Conditional-KG design and HACISU, we propose a simple construction model to benchmark HACISU. Experimental results show that our benchmark model outperforms several baselines but still has a considerable margin with human performance. We highlight the significance of HACISU, as it is the first carefully annotated dataset with conditional information. Our dataset is publicly available in http://101.200.120.155:5555/, hoping to serve as a challenging testbed and an ideal benchmark for Conditional-KG construction.
AB - Facts are conditionally established in most cases. However, current Knowledge Graph (KG) techniques only focus on the modeling and representations of facts, neglecting the presence of conditions, which are necessary to establish the validity of facts. In this paper, we propose Conditional Knowledge Graph (Conditional-KG), which employs a three-layer hierarchical network to incorporate both facts and conditions. To facilitate research on the automatic construction of Conditional-KG, we manually annotate an innovative large-scale dataset named HACISU. Based on the Conditional-KG design and HACISU, we propose a simple construction model to benchmark HACISU. Experimental results show that our benchmark model outperforms several baselines but still has a considerable margin with human performance. We highlight the significance of HACISU, as it is the first carefully annotated dataset with conditional information. Our dataset is publicly available in http://101.200.120.155:5555/, hoping to serve as a challenging testbed and an ideal benchmark for Conditional-KG construction.
KW - Conditional Knowledge Graph
KW - Knowledge representation
KW - Open information extraction
UR - https://www.scopus.com/pages/publications/85176943350
U2 - 10.1007/978-981-99-7224-1_16
DO - 10.1007/978-981-99-7224-1_16
M3 - 会议稿件
AN - SCOPUS:85176943350
SN - 9789819972234
T3 - Communications in Computer and Information Science
SP - 207
EP - 219
BT - Knowledge Graph and Semantic Computing
A2 - Wang, Haofen
A2 - Han, Xianpei
A2 - Liu, Ming
A2 - Cheng, Gong
A2 - Liu, Yongbin
A2 - Zhang, Ningyu
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 24 August 2023 through 27 August 2023
ER -