YAGO 4.5: A Logically Consistent Knowledge Base with a Rich and Informative Taxonomy

Keskeiset käsitteet

YAGO 4.5 extends the YAGO 4 knowledge base by carefully incorporating a large part of the Wikidata taxonomy, while respecting logical constraints and the distinction between classes and instances. This results in a new, logically consistent version of YAGO that adds a rich layer of informative classes to the existing knowledge base.

Tiivistelmä

The paper presents the design and implementation of YAGO 4.5, a new version of the YAGO knowledge base that aims to address the shortcomings of YAGO 4 by incorporating a richer taxonomy from Wikidata.
Key highlights:

Design Rationale:

The authors chose to use the taxonomy of Schema.org as the upper-level taxonomy, as it is concise, maintained by a W3C consortium, and finds applications beyond its original purpose.
The authors decided to carefully incorporate selected parts of the Wikidata taxonomy into the YAGO taxonomy, as the lack of such classes was a major criticism of YAGO 4.

Design Principles:

Prefer properties over class membership where possible.
Choose the property with fewer objects when there is a choice between a property and its inverse.
The upper taxonomy exists to define formal properties that will be populated.
The lower taxonomy exists to convey human-intelligible information about its instances in a non-redundant form.

Upper Taxonomy:

The authors start with the 8 top-level classes from Schema.org and add a class for fictional entities.
They manually create additional classes, such as yago:Award, yago:Gender, and yago:BeliefSystem, to define the ranges of properties.

Lower Taxonomy:

The authors map each class in the YAGO upper taxonomy to one or more classes in Wikidata.
They import the subclasses of the mapped Wikidata classes, while respecting various constraints to ensure logical consistency.
They exclude certain Wikidata classes, such as those related to linguistic objects, abstract objects, and scholarly articles.

Instances:

The authors handle the cases where Wikidata entities are both instances and classes by creating generic instances.
They model commercial products as instances rather than classes to attach properties like manufacturer and awards.

Properties and Constraints:

The authors manually define the properties for each YAGO class, keeping only those that are of general interest and exist in Wikidata.
They add SHACL constraints for maximum cardinality, literal patterns, and domain and range constraints.

The resulting YAGO 4.5 knowledge base contains 132M facts and is logically consistent, addressing the shortcomings of YAGO 4 while maintaining its strengths.

Tilastot

YAGO 4.5 contains 132M facts, compared to 343M facts in YAGO 4 and 500M facts in Wikidata.
YAGO 4.5 has 133k classes, compared to 10k classes in YAGO 4 and 2.8M classes in Wikidata.
YAGO 4.5 has 108 predicates, compared to 140 predicates in YAGO 4 and 11k predicates in Wikidata.
The dump size of YAGO 4.5 is 142GB, compared to 280GB for YAGO 4 and 766GB for Wikidata.

Lainaukset

None.

Tärkeimmät oivallukset

YAGO 4.5

by Fabian Sucha... klo arxiv.org 04-11-2024

https://arxiv.org/pdf/2308.11884.pdf

Syvällisempiä Kysymyksiä

How does the design of YAGO 4.5 compare to other top-level ontologies like Cyc, SUMO, and DOLCE in terms of the trade-offs between formal expressiveness, human-readability, and ease of maintenance?

In comparison to other top-level ontologies such as Cyc, SUMO, and DOLCE, the design of YAGO 4.5 focuses on a balance between formal expressiveness, human-readability, and ease of maintenance.

Formal Expressiveness: YAGO 4.5 maintains a high level of formal expressiveness by incorporating a rich taxonomy from Schema.org and selectively integrating parts of the Wikidata taxonomy. This allows for automated reasoning and logical consistency within the knowledge base.

Human-Readability: YAGO 4.5 emphasizes human-readability by ensuring that the taxonomy is concise and understandable. By mapping classes to Wikidata in a declarative manner, the ontology remains accessible to users who need to interpret and interact with the data.

Ease of Maintenance: The design principles of YAGO 4.5, such as preferring properties over class membership and focusing on a clean upper-level taxonomy, contribute to the ease of maintenance. By simplifying the taxonomy and enforcing constraints, the ontology becomes more manageable and less prone to inconsistencies.
In contrast, top-level ontologies like Cyc, SUMO, and DOLCE may prioritize formal expressiveness over human-readability, leading to complex and extensive hierarchies that can be challenging to maintain. YAGO 4.5 strikes a balance by prioritizing clarity and coherence while still maintaining a high level of formal expressiveness.

How does the design of YAGO 4.5 compare to other top-level ontologies like Cyc, SUMO, and DOLCE in terms of the trade-offs between formal expressiveness, human-readability, and ease of maintenance?

In comparison to other top-level ontologies such as Cyc, SUMO, and DOLCE, the design of YAGO 4.5 focuses on a balance between formal expressiveness, human-readability, and ease of maintenance.

Formal Expressiveness: YAGO 4.5 maintains a high level of formal expressiveness by incorporating a rich taxonomy from Schema.org and selectively integrating parts of the Wikidata taxonomy. This allows for automated reasoning and logical consistency within the knowledge base.

Human-Readability: YAGO 4.5 emphasizes human-readability by ensuring that the taxonomy is concise and understandable. By mapping classes to Wikidata in a declarative manner, the ontology remains accessible to users who need to interpret and interact with the data.

Ease of Maintenance: The design principles of YAGO 4.5, such as preferring properties over class membership and focusing on a clean upper-level taxonomy, contribute to the ease of maintenance. By simplifying the taxonomy and enforcing constraints, the ontology becomes more manageable and less prone to inconsistencies.
In contrast, top-level ontologies like Cyc, SUMO, and DOLCE may prioritize formal expressiveness over human-readability, leading to complex and extensive hierarchies that can be challenging to maintain. YAGO 4.5 strikes a balance by prioritizing clarity and coherence while still maintaining a high level of formal expressiveness.

How does the design of YAGO 4.5 compare to other top-level ontologies like Cyc, SUMO, and DOLCE in terms of the trade-offs between formal expressiveness, human-readability, and ease of maintenance?

In comparison to other top-level ontologies such as Cyc, SUMO, and DOLCE, the design of YAGO 4.5 focuses on a balance between formal expressiveness, human-readability, and ease of maintenance.

Formal Expressiveness: YAGO 4.5 maintains a high level of formal expressiveness by incorporating a rich taxonomy from Schema.org and selectively integrating parts of the Wikidata taxonomy. This allows for automated reasoning and logical consistency within the knowledge base.

Human-Readability: YAGO 4.5 emphasizes human-readability by ensuring that the taxonomy is concise and understandable. By mapping classes to Wikidata in a declarative manner, the ontology remains accessible to users who need to interpret and interact with the data.

Ease of Maintenance: The design principles of YAGO 4.5, such as preferring properties over class membership and focusing on a clean upper-level taxonomy, contribute to the ease of maintenance. By simplifying the taxonomy and enforcing constraints, the ontology becomes more manageable and less prone to inconsistencies.
In contrast, top-level ontologies like Cyc, SUMO, and DOLCE may prioritize formal expressiveness over human-readability, leading to complex and extensive hierarchies that can be challenging to maintain. YAGO 4.5 strikes a balance by prioritizing clarity and coherence while still maintaining a high level of formal expressiveness.