TY - GEN
T1 - Each Snapshot to Each Space
T2 - 21st International Semantic Web Conference, ISWC 2022
AU - Li, Yancong
AU - Zhang, Xiaoming
AU - Zhang, Bo
AU - Ren, Haiying
N1 - Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - Temporal knowledge graphs (TKGs) organize and manage the dynamic relations between entities over time. Inferring missing knowledge in TKGs, known as temporal knowledge graph completion (TKGC), has become an important research topic. Previous models handle all facts with different timestamps in an identical latent space, even though the semantic space of the TKG changes over time. Therefore, they are not effective to reflect the temporality of knowledge. To effectively learn the time-aware information of TKGs, different latent spaces are adapted for temporal snapshots at different timestamps, which yields a novel model, i.e., Space Adaptation Network (SANe). Specifically, we extend convolutional neural networks (CNN) to map the facts with different timestamps into different latent spaces, which can effectively reflect the dynamic variation of knowledge. Meanwhile, a time-aware parameter generator is designed to explore the overlap of latent spaces, which endows CNN with specific parameters in term of the context of timestamps. Therefore, knowledge in adjacent time intervals is efficiently shared to boost the performance of TKGC, which can learn the validity of knowledge over a period of time. Extensive experiments demonstrate that SANe achieves state-of-the-art performance on four well-established benchmark datasets for temporal knowledge graph completion.
AB - Temporal knowledge graphs (TKGs) organize and manage the dynamic relations between entities over time. Inferring missing knowledge in TKGs, known as temporal knowledge graph completion (TKGC), has become an important research topic. Previous models handle all facts with different timestamps in an identical latent space, even though the semantic space of the TKG changes over time. Therefore, they are not effective to reflect the temporality of knowledge. To effectively learn the time-aware information of TKGs, different latent spaces are adapted for temporal snapshots at different timestamps, which yields a novel model, i.e., Space Adaptation Network (SANe). Specifically, we extend convolutional neural networks (CNN) to map the facts with different timestamps into different latent spaces, which can effectively reflect the dynamic variation of knowledge. Meanwhile, a time-aware parameter generator is designed to explore the overlap of latent spaces, which endows CNN with specific parameters in term of the context of timestamps. Therefore, knowledge in adjacent time intervals is efficiently shared to boost the performance of TKGC, which can learn the validity of knowledge over a period of time. Extensive experiments demonstrate that SANe achieves state-of-the-art performance on four well-established benchmark datasets for temporal knowledge graph completion.
KW - Parameter generation
KW - Space adaptation
KW - Temporal knowledge graph
KW - Temporal knowledge graph completion
UR - https://www.scopus.com/pages/publications/85141669145
U2 - 10.1007/978-3-031-19433-7_15
DO - 10.1007/978-3-031-19433-7_15
M3 - 会议稿件
AN - SCOPUS:85141669145
SN - 9783031194320
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 248
EP - 266
BT - The Semantic Web – ISWC 2022 - 21st International Semantic Web Conference, Proceedings
A2 - Sattler, Ulrike
A2 - Hogan, Aidan
A2 - Keet, Maria
A2 - Presutti, Valentina
A2 - Almeida, João Paulo A.
A2 - Takeda, Hideaki
A2 - Monnin, Pierre
A2 - Pirrò, Giuseppe
A2 - d’Amato, Claudia
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 23 October 2022 through 27 October 2022
ER -