The YAGO Knowledge Base
CC-BY
Fabian M. Suchanek
>lm&kbs
>kbs&dbs
Language Models and Knowledge Bases
•
designed to generalize
(applies to machine learning models in general)
•
designed to memorize
(applies to structured data in general)
2
Language Models
Knowledge Bases
Language Models and Knowledge Bases
•
designed to generalize
(applies to machine learning models in general)
•
designed to memorize
(applies to structured data in general)
3
Language Models
Knowledge Bases
You give the blue dots
it generalizes
to a line
it stores the
two dots
Language Models and Knowledge Bases
•
designed to generalize
•
probabilistic
•
opaque
•
costly inference
•
creative/ flexible
=>
ideal for natural language processing
(applies to machine learning models in general)
•
designed to memorize
•
deterministic
•
transparent
•
cheap inference
•
“dumb” / rigid
=> ideal for storing crisp data
(applies to structured data in general)
4
Language Models
Knowledge Bases
Databases and Knowledge Bases
•
entities are stored in a table and
have exactly the attributes of table
•
fast on entity‐centric queries
•
has reasoning via constraints
•
not made for taxonomies
=>
ideal for complete data
and “flat” data
•
entities are stored in a graph and
can have (nearly) any attributes
•
need lots of joins for such queries
•
have good reasoning support
•
have a taxonomy, inherited properties
=>
ideal for incomplete data
and data with taxonomic hierarchy
5
Databases
Knowledge Bases
1
2
1 2 / /
5
6
Knowledge Base: Inheritance
Relations (properties, attributes) are declared for a class, and inherited by the subclasses.
Human
Entity
type
subclass
subclass
Singer
Elvis_Presley_(singer)
- label
- URL
- image
- birth date
- birth place
- spouse
- song
- awards
7
Knowledge Base: Terminology
A
Knowledge Base
(KB, also: Knowledge graph, KG) is a directed labeled graph, where the nodes
are entities and the edges are relations between these entities.
Human
Entity
type
subclass
subclass
Song
CreativeWork
subclass
subclass
type
sang
Taxonomy
of classes
Instances
with facts
Singer
Elvis_Presley_(singer)
Blue_Suede_Shoes_(song)
Labels
“Elvis”
“Blue Suede Shoes”
“Синие замшевые туфли”
label
label
- label
- URL
- image
Relation definitions
8
Knowledge Base: Terminology
A
Knowledge Base
(KB, also: Knowledge graph, KG) is a directed labeled graph, where the nodes
are entities and the edges are relations between these entities.
Human
Entity
type
subclass
subclass
Song
CreativeWork
subclass
subclass
type
sang
Taxonomy
of classes
Instances
with facts
Singer
Elvis_Presley_(singer)
Blue_Suede_Shoes_(song)
Labels
“Elvis”
label
label
Constraints:
Human⊓
CreativeWork
≡ ⊥
“Blue Suede Shoes”
“Синие замшевые туфли”
- label
- URL
- image
Relation definitions
9
Knowledge Base: Terminology
A
Knowledge Base
(KB, also: Knowledge graph, KG) is a directed labeled graph.
Human
Entity
type
subclass
subclass
Song
CreativeWork
subclass
subclass
type
sang
Taxonomy
of classes
Instances
with facts
Singer
Elvis_Presley_(singer)
Blue_Suede_Shoes_(song)
Labels
“Elvis”
label
label
Constraints:
Human⊓
CreativeWork
≡ ⊥
“Blue Suede Shoes”
“Синие замшевые туфли”
- label
- URL
- image
Relation definitions
Schema/ontology
10
[
Noy &al: Industry-Scale Knowledge Graphs — Lessons and Challenges, CACM 2019
]
= where I gave a talk
[Weikum, Dong, Razniewski, Suchanek: Machine Knowledge, Found. & Trends in Databases, 2021]
Knowledge Bases in Industry
11
[Weikum, Dong, Razniewski, Suchanek: Machine Knowledge, Found. & Trends in Databases, 2021]
Knowledge Bases in Industry: Construction