Scaling Ontology Reasoning via Individual Summarization

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 212 papers19 conceptsfrozen 2026-10-04

In the world of Description Logics (DL) and the Semantic Web, the "ABox" contains the ground facts of a knowledge base—the specific individuals and their relationships. As these datasets grow to include millions of records, traditional reasoners often hit a "memory wall." The computational cost of checking consistency or answering queries scales poorly with the number of individuals.

However, a recurring insight across recent research is that large datasets are often logically redundant. If two individuals share the same types and participate in the same kinds of relationships, a reasoner shouldn't need to process them separately. By collapsing these redundant entities into a single "proxy," we can reduce the size of the problem by several orders of magnitude without losing logical rigor.

The Intuition of Logical Redundancy

Imagine a database of a million students. From a logical perspective, if ten thousand of those students are only defined as "Undergraduates" enrolled in "Intro to Logic," they are indistinguishable for many reasoning tasks.

Achille Fokoue, Aaron Kershenbaum, and Li Ma demonstrated in SHIN ABox Reduction that we can distill these massive datasets down to their "logical essence." By identifying assertions that do not contribute to global logical propagation, they showed that consistency checking for millions of records can be reduced to a task small enough to fit in standard system memory.

Summarization via Proxy Individuals

The core mechanism of this research direction is summarization. Instead of reasoning over the raw data, we create an abstract model where one representative individual stands in for an entire equivalence class of similar individuals.

In the SHER system, Kavitha Srinivas introduced type-based individual summarization in OWL Reasoning in the Real World: Searching for Godot. The logic is straightforward: if the summary representation is logically consistent, the original, much larger dataset is guaranteed to be consistent as well. This "summary-first" approach allows systems to verify the integrity of massive enterprise-scale data by operating on a fraction of the original instances.

This concept extends to more complex structures. Michael Cochez and Patrick Koopmann use simulation graphs in Towards Conceptual Clustering in EL with Simulation Graphs to group individuals that satisfy identical logical concepts. This structural summarization can compress hundreds of thousands of individuals into a handful of nodes while preserving the essential logical relationships required for reasoning.

Pay-As-You-Go and Abstraction Refinement

A challenge with summarization is that it can be "too" aggressive, potentially hiding a logical conflict that only appears when individuals are treated separately. To solve this, researchers use a "pay-as-you-go" or refinement strategy.

The PAGOdA system, developed by Bernardo Cuenca Grau, Ian Horrocks, and colleagues in PAGOdA: Pay-As-You-Go Ontology Query Answering Using a Datalog Reasoner, uses an abstract summary to quickly invalidate "spurious" candidate answers. If the summary suggests a query might be satisfied, but the logic is ambiguous, the system only then "pays" the computational cost of full-scale reasoning on the relevant subset of data.

Similarly, Birte Glimm and Yevgeny Kazakov proposed in Ontology Materialization by Abstraction Refinement in Horn SHOIF that reasoning over abstract types can scale materialization tasks to datasets that would otherwise exceed time limits. If an abstraction is found to be too coarse, it is iteratively refined until the reasoning is complete.

Handling Equality and Query Optimization

Summarization also addresses the "clique" problem in logic. When a system learns that multiple names refer to the same entity (equality), it can create a massive web of redundant facts. Boris Motik and Ian Horrocks showed in Combining Rewriting and Incremental Materialisation Maintenance for Datalog Programs with Equality that collapsing these cliques into single canonical representatives significantly reduces storage and processing overhead.

Finally, these summaries serve as a powerful tool for query optimization. As Wafaa EL HUSSEINI and François Goasdoué describe in Query Optimization for Ontology-Mediated Query Answering, using quotient-based database summaries allows a system to prune irrelevant parts of a query before it even hits the database. By searching the "reduced space" of the summary first, they achieve speedups of up to three orders of magnitude.

By shifting the focus from individual instances to representative abstractions, this research direction transforms the "Big Data" problem in logic from a memory bottleneck into a manageable task of reasoning over essential structures.

Papers in this version

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 17 papers
  2. Version 1 · 2026-10-04 · 12 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_knowledge_base_summarization_616993bf,
  title = {knowledge base summarization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/knowledge-base-summarization/rdr_616993bfcf9c4c1098e65f77a8b6ca8e},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_scaling_ontology_reasoning_via_individua_616993bf_v1,
  title = {Scaling Ontology Reasoning via Individual Summarization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/scaling-ontology-reasoning-via-individual-summarization/rdr_616993bfcf9c4c1098e65f77a8b6ca8e/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}