Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Chain-of-History Reasoning for Temporal Knowledge Graph Forecasting

Yuwei Xia, Ding Wang, Qiang Liu, Liang Wang, Shu Wu, Xiaoyu Zhang

TL;DR

This work addresses the limitations of purely graph-based or single-step LLM-based TKG forecasting by introducing Chain-of-History (CoH) reasoning. CoH interrogates high-order histories step-by-step with an LLM to reveal semantically rich temporal patterns, while remaining a plug-and-play module that fuses LLM-derived insights with graph-based predictors via a simple score fusion rule. Empirical results on ICEWS datasets show that CoH improves both pure LLM predictions and, more notably, the predictive power of graph-based TKG models, with ablations confirming the value of high-order histories, stepwise inference, and proper candidate-scoring order. The approach highlights a practical path to integrate semantic reasoning into structured temporal graphs, balancing interpretability and performance, and points to future work on adaptive fusion and efficiency improvements.

Abstract

Temporal Knowledge Graph (TKG) forecasting aims to predict future facts based on given histories. Most recent graph-based models excel at capturing structural information within TKGs but lack semantic comprehension abilities. Nowadays, with the surge of LLMs, the LLM-based TKG prediction model has emerged. However, the existing LLM-based model exhibits three shortcomings: (1) It only focuses on the first-order history for prediction while ignoring high-order historical information, resulting in the provided information for LLMs being extremely limited. (2) LLMs struggle with optimal reasoning performance under heavy historical information loads. (3) For TKG prediction, the temporal reasoning capability of LLM alone is limited. To address the first two challenges, we propose Chain-of-History (CoH) reasoning which explores high-order histories step-by-step, achieving effective utilization of high-order historical information for LLMs on TKG prediction. To address the third issue, we design CoH as a plug-and-play module to enhance the performance of graph-based models for TKG prediction. Extensive experiments on three datasets and backbones demonstrate the effectiveness of CoH.

Chain-of-History Reasoning for Temporal Knowledge Graph Forecasting

TL;DR

This work addresses the limitations of purely graph-based or single-step LLM-based TKG forecasting by introducing Chain-of-History (CoH) reasoning. CoH interrogates high-order histories step-by-step with an LLM to reveal semantically rich temporal patterns, while remaining a plug-and-play module that fuses LLM-derived insights with graph-based predictors via a simple score fusion rule. Empirical results on ICEWS datasets show that CoH improves both pure LLM predictions and, more notably, the predictive power of graph-based TKG models, with ablations confirming the value of high-order histories, stepwise inference, and proper candidate-scoring order. The approach highlights a practical path to integrate semantic reasoning into structured temporal graphs, balancing interpretability and performance, and points to future work on adaptive fusion and efficiency improvements.

Abstract

Temporal Knowledge Graph (TKG) forecasting aims to predict future facts based on given histories. Most recent graph-based models excel at capturing structural information within TKGs but lack semantic comprehension abilities. Nowadays, with the surge of LLMs, the LLM-based TKG prediction model has emerged. However, the existing LLM-based model exhibits three shortcomings: (1) It only focuses on the first-order history for prediction while ignoring high-order historical information, resulting in the provided information for LLMs being extremely limited. (2) LLMs struggle with optimal reasoning performance under heavy historical information loads. (3) For TKG prediction, the temporal reasoning capability of LLM alone is limited. To address the first two challenges, we propose Chain-of-History (CoH) reasoning which explores high-order histories step-by-step, achieving effective utilization of high-order historical information for LLMs on TKG prediction. To address the third issue, we design CoH as a plug-and-play module to enhance the performance of graph-based models for TKG prediction. Extensive experiments on three datasets and backbones demonstrate the effectiveness of CoH.
Paper Structure (34 sections, 2 equations, 5 figures, 8 tables)

This paper contains 34 sections, 2 equations, 5 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Temporal Knowledge Graph Prediction.
  4. High-order History Chains in TKGs.
  5. Chain-of-History Reasoning over Temporal Knowledge Graph
  6. History Processing
  7. Reasoning Steps
  8. Step 1 to Step $k$-1 Reasoning.
  9. Step $k$ Reasoning.
  10. Results Processing and Fusion
  11. Experiments
  12. Experimental Settings
  13. Evaluation
  14. CoH Implementation Details
  15. Performance Comparison
  16. ...and 19 more sections

Figures (5)

  • Figure 1: An example of reasoning over TKG with LLMs. In Figures (a) and (b), we provide LLMs with different histories, which prompt LLMs to reason different answers for the predicted fact.
  • Figure 2: The performance (MRR (%)) of LLMs of two sizes based on different history lengths on TKG prediction. The provided histories contain both first- and second-order histories. The y-axis represents the MRR (%) value, and the x-axis denotes the total length of provided first- and second-order histories. The results are based on the commonly used TKG dataset ICEWS14.
  • Figure 3: An illustration of a two-step CoH reasoning procedure. In the first step, LLMs are provided with only first-order histories and asked to infer the most important histories. In the second step, LLMs are provided with second-order history chains based on the inferred first-order histories and asked to infer possible answers to the given query. Then the answers inferred by LLMs and graph-based models are adaptively fused to make the final prediction. Note that this only serves as a two-step reasoning example, more steps can be executed with CoH.
  • Figure 4: Performance of graph-based models plugged with CoH under different $\alpha$-values in terms of MRR (%). The x-axis denotes different $\alpha$-values, and the y-axis shows MRR (%) values.
  • Figure 5: Performance of graph-based models plugged with CoH under different $w$-values in terms of MRR (%). The x-axis denotes different $w$-values, and the y-axis shows MRR (%) values.