Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generating Explanations to Understand and Repair Embedding-based Entity Alignment

Xiaobin Tian, Zequn Sun, Wei Hu

TL;DR

This paper presents the first framework that can generate explanations for understanding and repairing embedding-based EA results by constructing an alignment dependency graph and resolving three types of alignment conflicts based on dependency graphs.

Abstract

Entity alignment (EA) seeks identical entities in different knowledge graphs, which is a long-standing task in the database research. Recent work leverages deep learning to embed entities in vector space and align them via nearest neighbor search. Although embedding-based EA has gained marked success in recent years, it lacks explanations for alignment decisions. In this paper, we present the first framework that can generate explanations for understanding and repairing embedding-based EA results. Given an EA pair produced by an embedding model, we first compare its neighbor entities and relations to build a matching subgraph as a local explanation. We then construct an alignment dependency graph to understand the pair from an abstract perspective. Finally, we repair the pair by resolving three types of alignment conflicts based on dependency graphs. Experiments on a variety of EA datasets demonstrate the effectiveness, generalization, and robustness of our framework in explaining and repairing embedding-based EA results.

Generating Explanations to Understand and Repair Embedding-based Entity Alignment

TL;DR

This paper presents the first framework that can generate explanations for understanding and repairing embedding-based EA results by constructing an alignment dependency graph and resolving three types of alignment conflicts based on dependency graphs.

Abstract

Entity alignment (EA) seeks identical entities in different knowledge graphs, which is a long-standing task in the database research. Recent work leverages deep learning to embed entities in vector space and align them via nearest neighbor search. Although embedding-based EA has gained marked success in recent years, it lacks explanations for alignment decisions. In this paper, we present the first framework that can generate explanations for understanding and repairing embedding-based EA results. Given an EA pair produced by an embedding model, we first compare its neighbor entities and relations to build a matching subgraph as a local explanation. We then construct an alignment dependency graph to understand the pair from an abstract perspective. Finally, we repair the pair by resolving three types of alignment conflicts based on dependency graphs. Experiments on a variety of EA datasets demonstrate the effectiveness, generalization, and robustness of our framework in explaining and repairing embedding-based EA results.
Paper Structure (36 sections, 10 equations, 6 figures, 8 tables, 2 algorithms)

This paper contains 36 sections, 10 equations, 6 figures, 8 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Notation
  4. Definitions
  5. Framework Overview
  6. EA Explanation and Dependency Construction
  7. Explanation Generation
  8. Alignment Dependency Graph Construction
  9. Entity Alignment Repair
  10. Relation Alignment Conflicts
  11. One-to-Many Conflicts
  12. Low-confidence Conflicts
  13. Experiments
  14. Setup
  15. Environment
  16. ...and 21 more sections

Figures (6)

  • Figure 1: Overview of the ExEA framework.
  • Figure 2: Illustration of explanation generation and ADG construction. In ADG, each edge has a weight indicating how a node can influence others. The edge weight is calculated based on the functionality of relations. For example, $0.759 = \min(\texttt{ifunc}(前任), \texttt{ifunc}(predecessor))$. The node in ADG denotes an EA pair, which has a match confidence calculated based on neighbor alignment. For example, $0.808 = \texttt{sigmod}(0.960\times0.759 + 0.937\times0.757)$.
  • Figure 3: Illustration of EA conflicts. The blue dotted arrow denotes the predicted "sameAs" relation of an EA model while the black dotted arrow denotes the inferred "$\neg$ sameAs" relation based on our method. $\beta$ is a predefined confidence threshold.
  • Figure 4: Time cost (s) of explanation generation for Dual-AMN on ZH-EN.
  • Figure 5: Case study of our explanations for EA models. The two dotted boxes in each subgraph represent the predicted aligned entities, which may be incorrect. The blue boxes and arrows form our explanation for the prediction.
  • ...and 1 more figures