figure 3: example queries for the YAGo knowledge base.

Sept. 2, 1945

July 28, 1914

<

$b

Germany

Politician

>

(bornin| livesin| citizenof) .locatedin*

*

Max Planck

Angela Merkel

isa

*

*

$x

hasWon

$x nobel Prize

isa

bornon

diedon $d

>

Jim Gray

fatherof

Scientist

$c

diedon

$y

KnowItAll and TextRunner are examples of Statistical-Web methods for large-scale knowledge acquisition.

YAGo for Large-Scale
Semantic Knowledge

Our YAGO project shares the

23, 24

KnowItAll and TextRunner goal of large-scale knowledge harvesting but emphasizes high accuracy and consistency rather than high recall ( coverage). YAGO is best characterized as a Semantic-Web approach, gathering its knowledge by (primarily) integrating information from Wikipedia and WordNet. It also employs text-mining-based techniques. YAGO contains close to two million entities and about 20 million facts about them, where facts are instances of binary relations. Extensive sampling has shown that YAGO accuracy is at least 95%, and many of its errors (false positives) are due to incorrect entries in Wikipedia itself. YAGO is publicly available at www.mpi-inf.mpg.de/yago/.

Two Wikipedia assets—infoboxes and the category system—are almost structured data. Infoboxes are collections of attribute name-value pairs often based on templates and reused for important types of entities (such as countries, companies, scientists, music bands, and sports teams). For example, the infobox for Max Planck delivers such data as birth_date = April 23, 1858, birth_place = Kiel, death_date = October 4, 1947, nationality = Germany, and alma_mater = Ludwig-Max-imilians-Universität München. As for the category system, the Max Planck article is manually placed in such categories as German_Nobel_laureates,

Nobel_laureates_in_physics, quantum_ physics, and University_of_Munich_ alumni. All give YAGO clues about instanceOf relations, so it can infer that the entity Max Planck is an instance of the classes GermanNobelLaureates, NobelLaureatesInPhysics, and Uni-versityOfMunichAlumni. But YAGO must be careful, as the placement in category quantum_physics does not mean that Max Planck is an instance of QuantumPhysics. The YAGO extractors employ linguistic processing (noun phrase parsing) and mapping rules to achieve high accuracy in harvesting the categories information.

These examples of YAGO information extraction indicate that relying solely on Wikipedia infoboxes and categories may result in a large but incoherent collection of facts. For example, we may know that Max Planck is an instance of GermanNobelLaureates but be unable to automatically infer that he is also an instance of Germans and of Nobel Laureates. Likewise, the fact that he was a physicist does not automatically tell us he was a scientist. To address these shortcomings, YAGO makes intensive use of the WordNet thesaurus (lightweight ontology), integrating the facts it harvests from Wikipedia with the taxonomic backbone provided by WordNet.

While WordNet knows many abstract classes and the “is-a” and “ part-of” relationships among them, it has only sparse information about individual entities that would populate its classes. The wealth of entities in Wikipedia complements WordNet nicely; conversely, the rigor and extensive coverage of WordNet’s taxonomy

compensate for the gaps and noise in the Wikipedia category system. Each individual entity YAGO discovers must be mapped into at least one existing YAGO class. If this fails, the entity and its related facts are not admitted into the knowledge base. Analogously, classes derived from Wikipedia category names (such as GermanNobelLaureates) must be mapped with a subclass relationship to one or more superclasses (such as NobelLaureates and Germans). These procedures ensure that YAGO maintains a consistent knowledge base, where consistency eliminates dangling entities and classes and guarantees that the subclass relation is acyclic.

Kylin/KOG. The “Intelligence in Wikipedia” project also extracts information from Wikipedia through its tools Kylin and Kylin Ontology

27

Generator (KOG). ²⁶ Whenever an infobox type includes an attribute in some articles but the attribute has no value for a given article, Kylin analyzes the full text of the article to derive the most likely value. Like KnowItAll and TextRunner (but unlike Libra, Cimple, and YAGO), Kylin pursues open extraction by considering all potentially significant attributes, even if they occur only sparsely in the entire Wikipedia corpus. KOG builds on Kylin’s output, unifies attribute names, derives type signatures, and (like YAGO) maps these entities onto the WordNet taxonomy through statistical relational learning. ¹⁵ KOG goes beyond YAGO by discovering new relationship types. It builds on the class system of both YAGO and DBpedia, ⁴ along with the entities in

62 communicAtionS of the Acm | APriL 2009 | voL. 52 | no. 4