Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HALO: Half Life-Based Outdated Fact Filtering in Temporal Knowledge Graphs

Feng Ding, Tingting Wang, Yupeng Gao, Shuo Yu, Jing Ren, Feng Xia

TL;DR

This work addresses the negative impact of outdated facts in temporal knowledge graphs (TKGs) on reasoning accuracy and training efficiency. It introduces HALO, a half-life-based outdated fact filtering framework consisting of three modules: a Temporal Fact Attention module to capture temporal evolution, a Dynamic Relation-aware Encoder to predict fact half-lives, and an Outdated Fact Filtering module using a time-decay mechanism to prune stale facts. Across ICEWS14, ICEWS18, and ICEWS05-15, HALO consistently improves state-of-the-art TKGR methods, validating its effectiveness in maintaining factual relevance. The results highlight the importance of accounting for fact expiration in temporal reasoning and suggest HALO’s potential for extending to multi-modal and multi-source knowledge graphs to further enhance robustness and efficiency.

Abstract

Outdated facts in temporal knowledge graphs (TKGs) result from exceeding the expiration date of facts, which negatively impact reasoning performance on TKGs. However, existing reasoning methods primarily focus on positive importance of historical facts, neglecting adverse effects of outdated facts. Besides, training on these outdated facts yields extra computational cost. To address these challenges, we propose an outdated fact filtering framework named HALO, which quantifies the temporal validity of historical facts by exploring the half-life theory to filter outdated facts in TKGs. HALO consists of three modules: the temporal fact attention module, the dynamic relation-aware encoder module, and the outdated fact filtering module. Firstly, the temporal fact attention module captures the evolution of historical facts over time to identify relevant facts. Secondly, the dynamic relation-aware encoder module is designed for efficiently predicting the half life of each fact. Finally, we construct a time decay function based on the half-life theory to quantify the temporal validity of facts and filter outdated facts. Experimental results show that HALO outperforms the state-of-the-art TKG reasoning methods on three public datasets, demonstrating its effectiveness in detecting and filtering outdated facts (Codes are available at https://github.com/yushuowiki/K-Half/tree/main ).

HALO: Half Life-Based Outdated Fact Filtering in Temporal Knowledge Graphs

TL;DR

This work addresses the negative impact of outdated facts in temporal knowledge graphs (TKGs) on reasoning accuracy and training efficiency. It introduces HALO, a half-life-based outdated fact filtering framework consisting of three modules: a Temporal Fact Attention module to capture temporal evolution, a Dynamic Relation-aware Encoder to predict fact half-lives, and an Outdated Fact Filtering module using a time-decay mechanism to prune stale facts. Across ICEWS14, ICEWS18, and ICEWS05-15, HALO consistently improves state-of-the-art TKGR methods, validating its effectiveness in maintaining factual relevance. The results highlight the importance of accounting for fact expiration in temporal reasoning and suggest HALO’s potential for extending to multi-modal and multi-source knowledge graphs to further enhance robustness and efficiency.

Abstract

Outdated facts in temporal knowledge graphs (TKGs) result from exceeding the expiration date of facts, which negatively impact reasoning performance on TKGs. However, existing reasoning methods primarily focus on positive importance of historical facts, neglecting adverse effects of outdated facts. Besides, training on these outdated facts yields extra computational cost. To address these challenges, we propose an outdated fact filtering framework named HALO, which quantifies the temporal validity of historical facts by exploring the half-life theory to filter outdated facts in TKGs. HALO consists of three modules: the temporal fact attention module, the dynamic relation-aware encoder module, and the outdated fact filtering module. Firstly, the temporal fact attention module captures the evolution of historical facts over time to identify relevant facts. Secondly, the dynamic relation-aware encoder module is designed for efficiently predicting the half life of each fact. Finally, we construct a time decay function based on the half-life theory to quantify the temporal validity of facts and filter outdated facts. Experimental results show that HALO outperforms the state-of-the-art TKG reasoning methods on three public datasets, demonstrating its effectiveness in detecting and filtering outdated facts (Codes are available at https://github.com/yushuowiki/K-Half/tree/main ).
Paper Structure (11 sections, 11 equations, 3 figures, 1 table)

This paper contains 11 sections, 11 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. HALO: An Outdated Fact Filtering Framework
  3. Temporal Fact Attention Module
  4. Dynamic Relation-aware Encoder Module
  5. Outdated Fact Filtering Module
  6. Fact Half-life Prediction
  7. Outdated Fact Filtering and Verification
  8. Experiments
  9. Results of TKG Reasoning
  10. Sensitivity Analysis
  11. Conclusion

Figures (3)

  • Figure 1: The changing temporal validity of facts within ICEWS14 dataset during the whole year of 2014.
  • Figure 2: Framework of HALO.
  • Figure 3: Sensitivity analysis on ICEWS18 and ICEWS14.