Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mixture Policy based Multi-Hop Reasoning over N-tuple Temporal Knowledge Graphs

Zhongni Hou, Miao Su, Xiaolong Jin, Zixuan Li, Long Bai, Jiafeng Guo, Xueqi Cheng

TL;DR

This work addresses the challenge of explainable, multi-hop reasoning over N-Tuple Temporal Knowledge Graphs (N-TKGs) by introducing MT-Path, a reinforcement learning framework that traverses historical n-tuples to predict future facts. MT-Path employs a mixture policy-driven action selector composed of three low-level policies—predicate-focused, core-element-focused, and whole-fact-focused—alongside an auxiliary element-aware GCN to capture temporal and semantic dependencies among facts. The model yields interpretable reasoning paths and demonstrates superior performance over static N-tuple and TKG/N-TKG baselines on NICE and NWIKI datasets, while enabling analysis of its reasoning process. Limitations include focusing on core-entity prediction and potential scalability challenges, with future work aiming to predict auxiliary entities and improve real-time applicability.

Abstract

Temporal Knowledge Graphs (TKGs), which utilize quadruples in the form of (subject, predicate, object, timestamp) to describe temporal facts, have attracted extensive attention. N-tuple TKGs (N-TKGs) further extend traditional TKGs by utilizing n-tuples to incorporate auxiliary elements alongside core elements (i.e., subject, predicate, and object) of facts, so as to represent them in a more fine-grained manner. Reasoning over N-TKGs aims to predict potential future facts based on historical ones. However, existing N-TKG reasoning methods often lack explainability due to their black-box nature. Therefore, we introduce a new Reinforcement Learning-based method, named MT-Path, which leverages the temporal information to traverse historical n-tuples and construct a temporal reasoning path. Specifically, in order to integrate the information encapsulated within n-tuples, i.e., the entity-irrelevant information within the predicate, the information about core elements, and the complete information about the entire n-tuples, MT-Path utilizes a mixture policy-driven action selector, which bases on three low-level policies, namely, the predicate-focused policy, the core-element-focused policy and the whole-fact-focused policy. Further, MT-Path utilizes an auxiliary element-aware GCN to capture the rich semantic dependencies among facts, thereby enabling the agent to gain a deep understanding of each n-tuple. Experimental results demonstrate the effectiveness and the explainability of MT-Path.

Mixture Policy based Multi-Hop Reasoning over N-tuple Temporal Knowledge Graphs

TL;DR

This work addresses the challenge of explainable, multi-hop reasoning over N-Tuple Temporal Knowledge Graphs (N-TKGs) by introducing MT-Path, a reinforcement learning framework that traverses historical n-tuples to predict future facts. MT-Path employs a mixture policy-driven action selector composed of three low-level policies—predicate-focused, core-element-focused, and whole-fact-focused—alongside an auxiliary element-aware GCN to capture temporal and semantic dependencies among facts. The model yields interpretable reasoning paths and demonstrates superior performance over static N-tuple and TKG/N-TKG baselines on NICE and NWIKI datasets, while enabling analysis of its reasoning process. Limitations include focusing on core-entity prediction and potential scalability challenges, with future work aiming to predict auxiliary entities and improve real-time applicability.

Abstract

Temporal Knowledge Graphs (TKGs), which utilize quadruples in the form of (subject, predicate, object, timestamp) to describe temporal facts, have attracted extensive attention. N-tuple TKGs (N-TKGs) further extend traditional TKGs by utilizing n-tuples to incorporate auxiliary elements alongside core elements (i.e., subject, predicate, and object) of facts, so as to represent them in a more fine-grained manner. Reasoning over N-TKGs aims to predict potential future facts based on historical ones. However, existing N-TKG reasoning methods often lack explainability due to their black-box nature. Therefore, we introduce a new Reinforcement Learning-based method, named MT-Path, which leverages the temporal information to traverse historical n-tuples and construct a temporal reasoning path. Specifically, in order to integrate the information encapsulated within n-tuples, i.e., the entity-irrelevant information within the predicate, the information about core elements, and the complete information about the entire n-tuples, MT-Path utilizes a mixture policy-driven action selector, which bases on three low-level policies, namely, the predicate-focused policy, the core-element-focused policy and the whole-fact-focused policy. Further, MT-Path utilizes an auxiliary element-aware GCN to capture the rich semantic dependencies among facts, thereby enabling the agent to gain a deep understanding of each n-tuple. Experimental results demonstrate the effectiveness and the explainability of MT-Path.
Paper Structure (15 sections, 11 equations, 3 figures, 4 tables)

This paper contains 15 sections, 11 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Methodology
  5. The Framework of Reinforcement Learning
  6. The Semantic Component
  7. The Mixture Policy-Driven Action Selector
  8. Optimization and Training.
  9. Experiments
  10. Experimental Setup
  11. Experimental Results
  12. Ablation Study
  13. Detailed Analysis
  14. Case study
  15. Conclusions

Figures (3)

  • Figure 1: The illustration of an n-tuple fact Barack Obama replaced George W. Bush as the 44th US president on 20 January 2009. This fact can be represented as Position Held (person: Barack Obama, position: President of US, replaces: George W. Bush, series ordinal: 44th, 2009/01/20).
  • Figure 2: An illustrative diagram of the proposed MT-Path model for multi-hop reasoning over N-TKGs. For the sake of brevity, each historical n-tuple's occurring time is not explicitly given.
  • Figure 3: Performance of MT-Path over queries involving seen and unseen entities.