Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Look Globally and Reason: Two-stage Path Reasoning over Sparse Knowledge Graphs

Saiping Guan, Jiyao Wei, Xiaolong Jin, Jiafeng Guo, Xueqi Cheng

TL;DR

LoGRe addresses sparse KG completion by abandoning external-assisted reasoning in favor of an inward, two-stage approach. Stage 1 builds a global relation-path reasoning schema from training data, separating type-specific and cross-type relations; Stage 2 aggregates paths from head entities to score candidate tails, with a similarity-based adjustment to improve reliability and explainability. Across five sparse KG benchmarks, LoGRe outperforms rule-based and path-based baselines and rivals embedding-based methods, with notable gains on NELL23K and strong explainability through explicit reasoning paths. The method also offers favorable complexity characteristics and demonstrates robust hyper-parameter behavior, suggesting practical applicability for interpretable reasoning in sparse knowledge graphs.

Abstract

Sparse Knowledge Graphs (KGs), frequently encountered in real-world applications, contain fewer facts in the form of (head entity, relation, tail entity) compared to more populated KGs. The sparse KG completion task, which reasons answers for given queries in the form of (head entity, relation, ?) for sparse KGs, is particularly challenging due to the necessity of reasoning missing facts based on limited facts. Path-based models, known for excellent explainability, are often employed for this task. However, existing path-based models typically rely on external models to fill in missing facts and subsequently perform path reasoning. This approach introduces unexplainable factors or necessitates meticulous rule design. In light of this, this paper proposes an alternative approach by looking inward instead of seeking external assistance. We introduce a two-stage path reasoning model called LoGRe (Look Globally and Reason) over sparse KGs. LoGRe constructs a relation-path reasoning schema by globally analyzing the training data to alleviate the sparseness problem. Based on this schema, LoGRe then aggregates paths to reason out answers. Experimental results on five benchmark sparse KG datasets demonstrate the effectiveness of the proposed LoGRe model.

Look Globally and Reason: Two-stage Path Reasoning over Sparse Knowledge Graphs

TL;DR

LoGRe addresses sparse KG completion by abandoning external-assisted reasoning in favor of an inward, two-stage approach. Stage 1 builds a global relation-path reasoning schema from training data, separating type-specific and cross-type relations; Stage 2 aggregates paths from head entities to score candidate tails, with a similarity-based adjustment to improve reliability and explainability. Across five sparse KG benchmarks, LoGRe outperforms rule-based and path-based baselines and rivals embedding-based methods, with notable gains on NELL23K and strong explainability through explicit reasoning paths. The method also offers favorable complexity characteristics and demonstrates robust hyper-parameter behavior, suggesting practical applicability for interpretable reasoning in sparse knowledge graphs.

Abstract

Sparse Knowledge Graphs (KGs), frequently encountered in real-world applications, contain fewer facts in the form of (head entity, relation, tail entity) compared to more populated KGs. The sparse KG completion task, which reasons answers for given queries in the form of (head entity, relation, ?) for sparse KGs, is particularly challenging due to the necessity of reasoning missing facts based on limited facts. Path-based models, known for excellent explainability, are often employed for this task. However, existing path-based models typically rely on external models to fill in missing facts and subsequently perform path reasoning. This approach introduces unexplainable factors or necessitates meticulous rule design. In light of this, this paper proposes an alternative approach by looking inward instead of seeking external assistance. We introduce a two-stage path reasoning model called LoGRe (Look Globally and Reason) over sparse KGs. LoGRe constructs a relation-path reasoning schema by globally analyzing the training data to alleviate the sparseness problem. Based on this schema, LoGRe then aggregates paths to reason out answers. Experimental results on five benchmark sparse KG datasets demonstrate the effectiveness of the proposed LoGRe model.
Paper Structure (28 sections, 9 equations, 3 figures, 7 tables)

This paper contains 28 sections, 9 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Related Works
  4. Embedding-based Models
  5. Rule-based Models
  6. Path-based Models
  7. Summarization and Comparison
  8. The Proposed LoGRe Model
  9. Stage1: Construction of Relation-path Reasoning Schema
  10. Type-specific Reasoning Schema
  11. Cross-type Reasoning Schema
  12. Relation-path Reasoning Schema
  13. Stage2: Path Reasoning
  14. The Reasoning Process
  15. The Scores of Candidate Tail Entities
  16. ...and 13 more sections

Figures (3)

  • Figure 1: The framework of the proposed LoGRe model.
  • Figure 2: A small example of the construction process for the relation-path reasoning schema. Dark red, yellow, and blue represent the entities of type person, organization, and geo-political entity, respectively.
  • Figure 3: Hyper-parameter analysis of LoGRe on WD-singer.

Theorems & Definitions (3)

  • Definition 1
  • Definition 2
  • Definition 3