Generalizing to Unseen Elements: A Survey on Knowledge Extrapolation for Knowledge Graphs

Mingyang Chen,Wen Zhang,Yuxia Geng,Zezhong Xu,Jeff Z. Pan,Huajun Chen

2023-12-16

Abstract:Knowledge graphs (KGs) have become valuable knowledge resources in various applications, and knowledge graph embedding (KGE) methods have garnered increasing attention in recent years. However, conventional KGE methods still face challenges when it comes to handling unseen entities or relations during model testing. To address this issue, much effort has been devoted to various fields of KGs. In this paper, we use a set of general terminologies to unify these methods and refer to them collectively as Knowledge Extrapolation. We comprehensively summarize these methods, classified by our proposed taxonomy, and describe their interrelationships. Additionally, we introduce benchmarks and provide comparisons of these methods based on aspects that are not captured by the taxonomy. Finally, we suggest potential directions for future research.

Computation and Language

What problem does this paper attempt to address?

The paper primarily focuses on a core issue in Knowledge Graphs (KGs): how to enable models to generalize to unseen entities or relations. Specifically, the paper attempts to address the following problems:

1. **Background Issue**: Traditional Knowledge Graph Embedding (KGE) methods face challenges when dealing with new entities or relations that appear during the testing phase. This is because traditional KGE methods assume that all entities and relations appearing in the training data will also appear during the testing phase, and that these entities and relations have already obtained suitable vector representations through training. However, in real-world applications, knowledge graphs are dynamic and constantly evolving, with new entities and relations emerging.

2. **Research Objective**: In response to the aforementioned background issue, this paper aims to conduct a comprehensive survey of methods proposed in recent years to address the generalization problem of unseen elements (entities or relations) in knowledge graphs. The authors collectively refer to these methods as "Knowledge Extrapolation" and propose a classification system to summarize these methods.

3. **Summary of Contributions**:
- Provides a common terminology set to unify relevant definitions.
- Summarizes existing knowledge extrapolation methods and categorizes them based on the proposed classification system.
- Describes the interrelationships between these methods.
- Introduces commonly used benchmark datasets and compares the performance of different methods.
- Finally, proposes potential directions for future research.

In summary, the main purpose of this paper is to provide a comprehensive and systematic review of the problem of handling unseen elements in the field of knowledge graphs, and to offer guidance and reference for subsequent research work.