DAT630 - Entity Retrieval I.

University of Stavanger, DAT630, 2016 Autumn

Transcript

1. DAT630
 Entity Retrieval I. Krisztian Balog | University of Stavanger 01/11/2016
2. Semantic Search
3. What is semantic search?
8. What is semantic search? - "Search with meaning" - Improve search accuracy by understanding searcher intent and the contextual meaning of terms/documents/… - Move beyond “ten blue links” (towards actually answering information needs) using rich context
9. Semantic search - Centers around entities - “Who was the first human in outer space?” - “How tall is the Eiffel tower?” - “Who is Brad Pitt married to?” - “Where is the closest Starbucks?” - “Which airlines fly the Airbus A380?” - “What is the best Chinese restaurant in Montreal?” - Entity/Attribute/Relationship retrieval - + social, + personal - + (hyper)local
10. Semantic search - Combination of entity-related techniques, 
 from various fields - Information Retrieval (IR) - Natural Language Processing (NLP) - Databases (DB) - Semantic Web (SW)
11. What is an entity?
12. people products organizations locations
13. What is an entity? - Uniquely identifiable thing or object - “A thing with a distinct and independent existence” - Characterized by having: - Unique ID - Name(s) - Type(s) - Attributes (/Descriptions) - Relationships to other entities
14. Entities… - are meaningful units for organizing information - are a key enabling component in semantic search
15. Entity Linking Here the aim is to identify the most significant topics; those which the document was written about (Maron, 1977). These index topics can be used to summarize the document and organize it under category-like headings. Wikipedia is a natural choice as a vocabulary for obtaining index topics, since it is broad enough to be applicable to most domains. To use Wikipedia in this way, one must go through much the same process as wikification: one must detect the significant terms being mentioned, and disambiguate these to the for training. For every link, a Wikipedian has manually—and probably with some effort—selected the correct destination to represent the intended sense of the anchor. This provides millions of manually-defined ground truth examples to learn from. All the experiments described in this paper are based on a version of Wikipedia that was released on November 20, 2007. It contains just under two million articles. Because we wanted a reasonable number of links to use for both training and evaluation, we selected articles containing at least 50 links. We also avoided lists Figure 1: A news story that has been automatically augmented with links to relevant Wikipedia articles Image taken from Milne and Witten (2008b). Learning to Link with Wikipedia. In CIKM '08.
16. Entity Retrieval
17. Entity Linking/Retrieval
18. Meet the Data
19. Data collection - Unstructured - Documents, web pages, snippets, … - Semistructured - XML, RDF, … - Structured - Relational DBs, RDF, … Often organized 
 around entities
20. Single most popular semistructured data source
21. Knowledge Bases - Aimed at machine understanding - Comprese a large set of large set of assertions about the world - Describe (specific) entities and their relationships
22. RDF Data Model - Resource Description Framework - Each resource is identified by a URI (Unique Resource Identifier) - (Entities = resources) - Assertions are represented as triples - Subject (resource) - Predicate (relation) - Object (resource or literal) subject object predicate
23. RDF Data Model subject object predicate Stavanger Norway locatedIn Stavanger hasPopulation 128369
24. Example - How can this information be represented using RDF? Kimi Räikkönen is a Finnish racing driver, born on October 17, 1979, currently driving for Ferrari in Formula One.
25. Example
26. Knowledge bases - Conceptually form a large, directed graph - Also called knowledge graphs when the emphasis is on relationships between entities <dbr:Kimi_Räikkönen> <dbr:Finland> <dbc:Ferrari_Formula_One_drivers> <dbr:2007_Belgian_Grand_Prix> <dbo:RacingDriver> "Räikkönen, Kimi" "1979-10-17" <foaf:name> <dbo:birthDate> <dbp:firstDriver> <dbo:nationality> <dct:subject> <rdf:type>
27. SPARQL - Structured query language for RDF SELECT ?p WHERE { ?p has-profession Computer_Scientist . ?p has-gender Female . ?p occurs-with "semantic search"}
28. Early Attempt: Cyc - Started in 1984 with the goal to manually build a knowledge base of everyday common knowledge - … still building and far from complete - "one of the most controversial endeavors of the artificial intelligence history"
29. - DBpedia - Freebase - Wikidata Popular (Public) Knowledge Bases
30. DBpedia - "A database version of Wikipedia" - Extracts RDF statements from Wikipedia articles - Mostly relies on infoboxes - Further homogenization or "normalization" is performed to achieve high data quality - Using manual mappings against the DBpedia Ontology
31. DBpedia Ontology Person Athlete MotorsportRacer RacingDriver FormulaOneRacer xsd:date Agent xsd:integer birthDate Place <owl:Thing> birthPlace age Event GrandPrix SportsEvent SocietalEvent firstRace firstWin SportsTeam currentTeam Organisation ceo FormulaOneTeam playerInTeam formationDate roleInEvent xsd:nonNegativeInteger wins rdf:langString champion medalist distanceLaps firstDriverTeam birthName
32. dbpedia.org
34. Linking Open Data (LOD)
35. (re)Branding - Semantic Web data - Linking Open Data - Web of Data
36. Proprietary Knowledge Bases Knowledge Graph Entity Graph Satori … the knowledge graph is one of Google's biggest search milestones of the last decade… —Amit Singhal, Google’s director of search See: https://www.youtube.com/watch?v=mmQl6VGvX-c
37. RDFa, Microdata, … - Different protocols for marking up web pages - schema.org - shared vocabulary - used by Google, Bing, Yandex, etc. - powers rich result snippets
38. Entity Retrieval
39. Entity retrieval Addressing information needs that are better answered by returning specific objects (entities) instead of just any type of documents.
40. 6 % 36 % 1 % 5 % 12 % 41 % Entity (“1978 cj5 jeep”) Type (“doctors in barcelona”) Attribute (“zip code waterville Maine”) Relation (“tom cruise katie holmes”) Other (“nightlife in Barcelona”) Uninterpretable Distribution of web search queries (Pound et al., 2010) Pound, Mika, and Zaragoza (2010). Ad-hoc object retrieval in the web of data. In WWW ’10.
41. 28 % 15 % 10 % 4 % 14 % 29 % Entity Entity+refiner Category Category+refiner Other Website Distribution of web search queries (Lin et al., 2011) Lin, Pantel, Gamon, Kannan, and Fuxman (2012). Active objects. In WWW ’12.
42. Entities - Objects (or "things") with - Unique identifier - Name(s) - Attributes and/or description - Type(s) - Relationships to other entities
43. Ranking Entities with Ready-made Descriptions
44. "Entity homepages"
45. Document-based entity representations - Most entities have a “home page” - I.e., each entity is described by a document - In this scenario, ranking entities is much like ranking documents - unstructured - semi-structured
46. Using Language Models - Standard document retrieval methods applied on entity description documents - Just replacing d with e P(e|q) / P(e)P(q|✓e ) = P(e) Y t2q P(t|✓e )n(t,q) Entity prior
 Probability of the entity 
 being relevant to any query Entity language model
 Multinomial probability distribution over the vocabulary of terms
47. Semi-structured entity representation - Entity description documents are rarely unstructured - Different sections, fields, etc.
49. How to rank entities in knowledge bases?
50. foaf:name Audi A4 rdfs:label Audi A4 rdfs:comment The Audi A4 is a compact executive car produced since late 1994 by the German car manufacturer Audi, a subsidiary of the Volkswagen Group. The A4 has been built [...] dbpprop:production 1994 2001 2005 2008 rdf:type dbpedia-owl:MeanOfTransportation dbpedia-owl:Automobile dbpedia-owl:manufacturer dbpedia:Audi dbpedia-owl:class dbpedia:Compact_executive_car owl:sameAs freebase:Audi A4 is dbpedia-owl:predecessor of dbpedia:Audi_A5 is dbpprop:similar of dbpedia:Cadillac_BLS dbpedia:Audi_A4
51. How to rank entities in knowledge bases? - Represent entities as fielded documents
52. Fielded models - Fielded extensions of document retrieval methods - E.g., Mixture of Language Models (MLM) m X j=1 µj = 1 Field language model
 Smoothed with a collection model built
 from all document representations of the
 same type in the collection Field weights P(t|✓d ) = m X j=1 µjP(t|✓dj )
53. Setting field weights - Heuristically - Proportional to the length of text content in that field, to the field’s individual performance, etc. - Empirically (using training queries) - Problems - Number of possible fields is huge - It is not possible to optimise their weights directly - Entities are sparse w.r.t. different fields - Most entities have only a handful of predicates
54. Predicate folding - Idea: reduce the number of fields by grouping them together - Grouping based on - type - manually determined importance
55. Predicate Folding
    foaf:name Audi A4 rdfs:label Audi A4 rdfs:comment The Audi A4 is a compact executive car produced since late 1994 by the German car manufacturer Audi, a subsidiary of the Volkswagen Group. The A4 has been built [...] dbpprop:production 1994 2001 2005 2008 rdf:type dbpedia-owl:MeanOfTransportation dbpedia-owl:Automobile dbpedia-owl:manufacturer dbpedia:Audi dbpedia-owl:class dbpedia:Compact_executive_car owl:sameAs freebase:Audi A4 is dbpedia-owl:predecessor of dbpedia:Audi_A5 is dbpprop:similar of dbpedia:Cadillac_BLS Name
 Attributes
 Out-relations
 In-relations

56. Entity Resolution
    foaf:name Audi A4 rdfs:label Audi A4 rdfs:comment The Audi A4 is a compact executive car produced since late 1994 by the German car manufacturer Audi, a subsidiary of the Volkswagen Group. The A4 has been built [...] dbpprop:production 1994 2001 2005 2008 rdf:type dbpedia-owl:MeanOfTransportation dbpedia-owl:Automobile dbpedia-owl:manufacturer dbpedia:Audi dbpedia-owl:class dbpedia:Compact_executive_car owl:sameAs freebase:Audi A4 is dbpedia-owl:predecessor of dbpedia:Audi_A5 is dbpprop:similar of dbpedia:Cadillac_BLS Name
 Attributes
 Out-relations
 In-relations
 - Need to replace entity URIs with their names - so that they become "searchable" terms Mean of transportation Audi A5