A Survey on Extractive Knowledge Graph Summarization: Applications, Approaches, Evaluation, and Future Directions

TL;DR

This survey addresses the challenge of distilling large, heterogeneous knowledge graphs into compact, faithful extractive summaries that preserve essential content for downstream tasks. It develops a two-tier taxonomy separating static and dynamic summaries, and systematically reviews methods that target pattern coverage (Class/Property, Characteristic Set, EDP, LP, Path) as well as answer coverage. It covers dynamic, query-biased formulations based on Group Steiner Tree variants (GST, DCGST, QGST) and personalization (single-entity, domain, query history), and it discusses evaluation frameworks, including the BANDAR benchmark and extrinsic task assessments. The paper highlights future directions toward neural, supervised, generative, comparative, and collaborative extraction to enhance scalability, adaptability, and cross-KG comparability in real-world KG applications.

Abstract

With the continuous growth of large Knowledge Graphs (KGs), extractive KG summarization becomes a trending task. Aiming at distilling a compact subgraph with condensed information, it facilitates various downstream KG-based tasks. In this survey paper, we are among the first to provide a systematic overview of its applications and define a taxonomy for existing methods from its interdisciplinary studies. Future directions are also laid out based on our extensive and comparative review.

Figures (13)

• Figure 1: A small KG as our running example.
• Figure 2: A taxonomy of methods for extractive KG summarization.
• Figure 3: Expressivity increases among data patterns.
• Figure 4: A summary extracted from Figure \ref{['fig:example']} containing the most frequently instantiated class Person and property act_in.
• Figure 5: Two EDPs ($E_1, E_2$) and a LP ($L$) in Figure \ref{['fig:example']}.