logo
CanyamAI summaries for academic research
language
MobileMobile
  • Phone FeatureWeChat Mini Program
EnglishEnglish
noticeNotice
BlogBlog
delete
arrow
Return

Time-enhanced compound geometric operations for temporal knowledge graph embedding

delete2025-08-12
delete0
PRE
AI
W
Wenhao Li
张
张东 (Dong Zhang)
G
Guanyu Li *
Y
Y. L. Zou
DOI：10.1016/j.neucom.2025.131182delete
deleteOriginal
deleteOriginal request for help
deleteShare
deleteSave
Abstract

Abstract

En 中文
Temporal knowledge graph embedding aims to represent entities, relations, and timestamps in a continuous vector space while preserving their temporal evolutionary patterns. TCompoundE is a recently proposed temporal knowledge graph embedding model that uses compound operations involving translation and scaling as the relation-specific and time-specific operations. However, this model has the following three problems: (1) The model only employs traditional geometric transformations and does not explore more complex transformation relationships. (2) The model fails to enhance the head entity with time-awareness, which weakens the model’s ability to model dynamic changes. (3) The model neglects the influence of the head entity and the relation in the modeling process. We propose the TTComE, a time-enhanced compound geometric temporal knowledge graph embedding model, to tackle these questions. Here are three improvement points: (1) We introduced the Spatiotemporal Rotation Operation for modeling, which enables the model to better capture complex relationships and temporal evolution patterns. (2) We performed the Temporal-aware Enhancement Operation on the head entities and relations respectively, to enhance the model’s ability to capture dynamic changes. (3) We introduced the Weight-adaptive Translation Operation, which assigns learnable weights to both the head entity and the relation, followed by a translation operation. This enables the model to adaptively adjust the contributions of the head entity and the relation in prediction tasks. Finally, we experimentally validated the model on multiple datasets, complemented by ablation studies, and the results demonstrated significant improvements over other baseline models.
Keywords:
Temporal knowledge graph embedding
Spatiotemporal Rotation Operation
Temporal-aware Enhancement
Weight-adaptive Translation
Dynamic change modeling

Journal

Neurocomputing cover
Neurocomputing
IF:
6.5
Papers:
2.5W
Citations:
6.5W

Organization

No organization information available