Ontology-based User Modeling and Recommendation. Iván Cantador Information Retrieval Group Universidad Autónoma de Madrid, Spain

Transcription

1 Ontology-based User Modeling and Recommendation Iván Cantador Information Retrieval Group Universidad Autónoma de Madrid, Spain

2 Contents 1 The Social Semantic Web Web 2.0: TheSocial Web Web 3.0: The Semantic Web Ontology-based User Modeling Ontology-based User Preferences Mapping Social Tags to Ontology Concepts Mapping Facebook Likes to DBpedia Entities Ontology-based Recommendation An Example of Ontology-based Recommender System A Linked Data-based Cross-domain Recommendation Approach Conclusions

4 Web Is it so easy?

5 4

6 5 Web 2.0 Web 1.0 = user generated contents, user generated contents,... and user generated contents

7 User generated contents 6 Web 2.0 = The Social Web Creation of diverse formats of user generated contents People: - Communicate online with contacts Social networks - Facebook, Twitter, LinkedIn,... - Upload or create multimedia objects - Maintain personal bookmarks - Post comments, reviews and ratings - Annotate resources

8 User generated contents 7 Web 2.0 = The Social Web Creation of diverse formats of user generated contents People: - Communicate online with contacts - Contribute to wiki-style repositories Wikipedia, AboutUs, wikihow, WikiMapia,... - Maintain personal bookmarks - Post comments, reviews and ratings - Annotate resources

9 User generated contents 8 Web 2.0 = The Social Web Creation of diverse formats of user generated contents People: - Communicate online with contacts - Contribute to wiki-style repositories - Upload or create multimedia objects YouTube, Flickr, Last.fm,... - A - A

10 User generated contents 9 Web 2.0 = The Social Web Creation of diverse formats of user generated contents People: - Communicate online with contacts - Contribute to wiki-style repositories - Upload or create multimedia objects - Maintain personal bookmarks Delicious, Digg, CiteULike,... - Annotate resources

11 User generated contents 10 Web 2.0 = The Social Web Creation of diverse formats of user generated contents People: - Communicate online with contacts - Contribute to wiki-style repositories - Upload or create multimedia objects - Maintain personal bookmarks - Post comments, reviews and ratings Amazon, Epinions, Blogger, Rotten Tomatoes,... Blogs and recommender systems

12 User generated contents 11 Web 2.0 = The Social Web Creation of diverse formats of user generated contents People: - Communicate online with contacts - Contribute to wiki-style repositories - Upload or create multimedia objects - Maintain personal bookmarks - Post comments, reviews and ratings - Annotate resources Social tagging systems

13 Web Bard, M The Big Book of Social Media: Case Studies, Stories, Perspectives.

14 13 Ok, the Web 2.0 is out there! What s next!?

16 Web

17 Web 3.0!? 16 Spivack, N Web Evolution.

18 Web 3.0 Let s say Web 2.5 ;-) 17 Pileggi, S.; Amor, R Addressing Semantic Geographic Information Systems. Future Internet 5, pp. 585

19 It s all about semantics 18 Semantics Bags of words: uncategorized terms e.g. a folksonomy

20 It s all about semantics 19 Taxonomies: categories + hierarchical relations e.g. Linnaen taxonomy Semantics human:=[kingdom: Animalia, class: Mammalia, order: Primates, family: Hominidae, genus: Homo, species: Homo sapiens] Bags of words: uncategorized terms e.g. a folksonomy

21 It s all about semantics 20 Thesauri: categories + fixed hierarchical & associative relations e.g. WordNet thesaurus human: S:(n) homo, man, human being, human(any living or extinct member of the family Hominidae characterized by superior intelligence, articulate speech, and erect carriage) direct hypernym: S:(n) hominid(a primate of the family Hominidae direct hyponym: S:(n) Homo sapiens(the only surviving hominid Taxonomies: categories + hierarchical relations e.g. Linnaen taxonomy Semantics human:=[kingdom: Animalia, class: Mammalia, order: Primates, family: Hominidae, genus: Homo, species: Homo sapiens] Bags of words: uncategorized terms e.g. a folksonomy

22 It s all about semantics 21 Ontologies: classes + instances + arbitrary semantic relations + rules e.g. DBpedia ontology Thesauri: categories + fixed hierarchical & associative relations e.g. WordNet thesaurus human: S:(n) homo, man, human being, human(any living or extinct member of the family Hominidae characterized by superior intelligence, articulate speech, and erect carriage) direct hypernym: S:(n) hominid(a primate of the family Hominidae direct hyponym: S:(n) Homo sapiens(the only surviving hominid Taxonomies: categories + hierarchical relations e.g. Linnaen taxonomy Semantics human:=[kingdom: Animalia, class: Mammalia, order: Primates, family: Hominidae, genus: Homo, species: Homo sapiens] Bags of words: uncategorized terms e.g. a folksonomy

23 Ontologies An ontology is a formal, explicitspecification of a sharedconceptualization Formal: machine-readable 22 Explicit: concepts, properties, relations, functions, constraints, axioms are explicitly defined Shared: consensual knowledge Conceptualization: abstract model and simplified vide of some phenomenon in the world that we want to represent Gruber, T Toward Principles for the Design of Ontologies Used for Knowledge Sharing. International Journal of Human-Computer Studies 43(5-6), pp

24 Ontologies and knowledge bases How to model the world of pizzas? ontology knowledge base 23 Classes Properties Instances Cuattro Formaggi hascountryorigin Restrictions Italy

25 Ontologies and knowledge bases 24 Querying data (SPARQL) Defining ontologies (RDFS, OWL) Expressing and linking metadata (RDF) Assigning unambiguous names (URI)

26 RDF (Resource Description Framework) RDF identifies things using Web identifiers (URIs), and describes resources with properties and property values. The triple representation (subject, predicate, object) 25 personnel/ivancantador 'Ivan' <?xml version="1.0"?> <rdf:rdf xmlns:rdf=" xmlns:epsp= <rdf:description rdf:about=" <epsp:name>ivan</epsp:name> <epsp:lastname>cantador</epsp:lastname> <epsp:nationality>spanish</epsp:nationality> <rdf:type>associate Professor</rdf:type> </rdf:description> </rdf:rdf>

27 RDFS (RDF Schema) 26 RDFS provides the framework to describe classes and properties, allowing the creation of hierarchies <?xml version="1.0"?> <rdf:rdf xmlns:rdf=" xmlns:rdfs=" xml:base= <rdfs:class rdf:id= Professor" /> <rdfs:class rdf:id= Associate Professor"> <rdfs:subclassof rdf:resource= #Professor"/> </rdfs:class> <rdf:propertyrdf:id= teachessubject > <rdfs:domain rdf:resource= #Professor"/> <rdfs:range rdf:resource= #Subject"/> </rdf:property> </rdf:rdf>

28 OWL (Web Ontology Language) OWL is a stronger language with greater machine interpretability than RDF/RDFS (reasoning support) OWL Little / OWL DL / OWL Full <?xml version="1.0"?> <rdf:rdf xmlns:rdf=" xmlns:rdfs= xmlns:owl = " xml:base= <owl:class rdf:id= Professor > <owl:restriction> <owl:onproperty rdf:resource= #hasacademictitle"/> <owl:hasvalue> PhD^^ </owl:hasvalue> <owl:restriction> </owl:class> </rdf:rdf> Iván has a PhD -> therefore Iván can be a professor! Value constraints: owl:allvaluesfrom owl:somevaluesfrom owl:hasvalue Cardinality constraints: owl:cardinality owl:mincardinality owl:maxcardinality. 27

29 SPARQL (SPARQL Protocol and RDF Query Language) SPARQL is a query language for RDF Based on the triple representation (subject, predicate, object) SPARQL 1.1 is W3C Recommendation since 21 st March 2013 PREFIX epsp: < <rdf:rdf xmlns:rdf=" SELECT?x?name WHERE {?x rdf:type epsp:associate Professor.?x epsp:name?name. Can you google this??x epsp:nationality Spanish } SPARQL lets the construction of very powerful queries - Filtering elements - Querying named graphs -Ordering by/ distinct / reduced / offset / limit -Count / sum / avg/ min / max / GroupConcat 28

30 Towards the Web Explicit semantics in the Web Microformats RDFa HTML5 microdata Schema.org Linked Open Data

31 Microformats 30 Microformatsuse existing HTML attributesto embed structured data types in an HTML document Expressive power is limited as microformatsare only designed to pre-defined vocabularies No interlinking between entities <div class="vcard"> <a class="fn org url" href=" <div class="adr"> <span class="street-address">francisco Tomas y Valiente</span > <span class="locality">madrid</span>, <span class="postal-code">28049</span> <span class="country-name">spain</span > </div> </div>

32 RDFa (Resource Description Framework in Attributes) 31 W3C standard for embedding RDF data in HTML documents A set of new HTML attributesandsspecs of how to use them RDFais just a syntax, the publisher has to choose the vocabulary RDFa is domain independent <html> <body vocab=" <h2 property= name >Ivan Cantador</h2> <p><span property= gender">undefined</span></p>... </body> </html> Web standard since June

33 RDFa (Resource Description Framework in Attributes) 32 Used by Facebook! among others RDF vocabulary used in conjunction with RDFa <html xmlns:og=" <head> <title>the Rock (1996)</title> <meta property="og:title" content="the Rock" /> <meta property="og:type" content="movie" /> <meta property="og:url" content=" /> <meta property="og:image" content=" /> </head>... </html>

34 HTML5 microdata HTML5 introduces inline elements (text-level semantics) to describe specific type of information <div itemscopeitemtype=" My name is <span itemprop="name">ivan</span> Here is my home page: <a href=" itemprop="url"> </a> I live in Madrid, and work as an <span itemprop="title >Associate Professor</span> at <span itemprop="affiliation">uam</span>. </div> Itemid Itemprop Itemref Itemscope itemtype 33

35 Schema.org Schema.org provides a collection of vocabularies and a micro-data format, which are recognized by the major search engine providers (de facto standard) <div itemscopeitemtype= <h1 itemprop="name">pirates of the Carribean: On Stranger Tides (2011)</h1> Director: <div itemprop="director" itemscope itemtype=" <span itemprop="name">rob Marshall</span> </div> 34

36 35 Linked Open Data (August 2014) 22nd January 2015, Madrid, Spain

37 Linked Open Data DBpedia: The Wikipedia ontology and knowledge base DBpedia, 36 Wikipedia

38 Linked Open Data FOAF: The Friend-Of-A-Friend ontology 37

40 User Models 39 What data from a user could be modeled, gathered, learned, and exploited? Among other things Musial, K Recommender System for Online Social Network. Lambert Academic Publishing.

41 Ontologies in User Modeling 40 Razmerita, L., Angehrn, A., Maedche, A Ontology-Based User Modeling for Knowledge Management Systems. In Proceedings of the 9th International Conference on User Modeling (UM 2003), pp

42 Ontologies in User Modeling 41 Cantador, I Exploiting the Conceptual Space in Hybrid Recommender Systems: A Semantic-based Approach. PhD thesis, Universidad Atónoma de Madrid.

43 Ontology-based User Preferences User and item profiles are represented as vectors u m =(u m,1,, u m,k ) [-1,+1] K and i n =(i n,1,, i n,k ) [0,+1] K, where u m,k, i n,k are the weights that measure the relevance of concept c k for user u m and item i n Users (M) 42 User Profile Weighted Semantic Preferences Ontology concepts (K) Weighted Semantic Annotations Items (N) Item Description

44 Ontology-based User Preferences C Art Work C Literary Work C Essay C Novel C Poem C Plastic Art Work C Architectural Work C Sculpture C Performing Art Work C Dance C Musical Composition C Theatre Play C Visual Art Work C Movie C Painting C Photograph C Institution C Cultural Institution C Library C Museum I El Prado C Economic Institution C Educational Institution C College C Institute of Technology C University C School C Family Institution C Medical Institution C Political Institution C Religious Institution C Social Welfare Institution C Region C C Geographical Region Political Region C C C C Continent Country City I State C I Madrid Class Instance subclassof instanceof 43

45 Ontology-based User Preferences C Art Work C Literary Work C Essay C Novel C Poem C Plastic Art Work C Architectural Work C Sculpture C Performing Art Work C Dance C Musical Composition C Theatre Play C Visual Art Work C Movie C Painting C Photograph C Institution C Cultural Institution C Library C Museum I El Prado C Economic Institution C Educational Institution C College C Institute of Technology C University C School C Family Institution C Medical Institution C Political Institution C Religious Institution C Social Welfare Institution C Region C C Geographical Region Political Region C C C C Continent Country City I State C I Madrid Class Instance subclassof instanceof 44

46 Ontology-based User Preferences C Art Work C Institution C C Literary Work C Cultural Institution C Essay C Novel C Poem C Plastic Art Work exhibitedat C Architectural Work C Sculpture C Performing Art Work C Dance C Musical Composition C Theatre Play C Visual Art Work C Movie C Painting C Photograph C Library C Museum I El Prado C Economic Institution C Educational Institution C College C Institute of Technology C University C School C Family Institution C Medical Institution C Political Institution C Religious Institution C Social Welfare Institution Region C C Geographical Region Political Region C C C C Continent Country City I State C I Madrid Class Instance subclassof instanceof 45

47 Ontology-based User Preferences C Art Work C Institution C C Literary Work C Cultural Institution C Essay C Novel C Poem C Plastic Art Work exhibitedat C Library C Museum I El Prado C Economic Institution locatedin C Architectural Work C Educational Institution C Sculpture C College C Performing Art Work C Institute of Technology C Dance C University C Musical Composition C School C Theatre Play C Family Institution C Visual Art Work C Medical Institution C Movie C Political Institution C Painting C Religious Institution C Photograph C Social Welfare Institution Region C C Geographical Region Political Region C C C C Continent Country City I State C I Madrid Class Instance subclassof instanceof 46

49 Social tagging 48 In social tagging systems, usersupload or create content (items), annotate it with freely chosen words (tags), and share these annotations with others The nature of tagged items is manifold: photos (Flickr) music tracks (Last.fm) video clips (YouTube) movies (MovieLens) web pages (Delicious) scientific articles (CiteULike, BibSonomy)...

50 Social tagging 49 Social tagging is a practice of manually creating and managing tags to annotate and categorize content This practice is also known as social classification, social indexing, and collaborative tagging Folksonomy, a term coined by Thomas Vander Walfrom a mailing list conversation with Gene Smith, is a portmanteau of folk and taxonomy Smith, G Folksonomy: Social Classification Vander Wal, T Folksonomy Coinage and Definition

51 Social tagging 50 Delicious, users who tagged the item tagged item (web page) tags assigned to the item

52 Social tagging 51 Delicious, tag-based search Room for improvement! non-relevant results

53 Social tagging Tags are free text, and thus suffer from various vocabulary problems Polysemy: bridge, jaguar, java Synonym: biscuit, cookie Morphological deviations: blog, blogs, blogging Multilinguality: spain, españa, spagna, espagne Contemporaneous terminology: podcast, ipod, diy Misspellings: barclona (instead of barcelona) Compound nouns: new_york, thelordoftheringsmovie Subjective/personal concepts: funny, my wife, to read 52 Room for improvement! Mathes, A Folksonomies-Cooperative Classification and Communication through Shared Metadata. Computer Mediated Communication -LIS590CMC, Graduate School of Library and Information Science, University of Illinois Urbana-Champaign, IL, USA.

54 Social tagging 53 To address the above problems, folskonomy-based information retrieval and filtering systems have to identify and exploit the underlying semantics of the tags

55 Purpose-based categorization of social tags Let s tag this Flickrphoto! 54

56 Purpose-based categorization of social tags Let s tag this Flickrphoto! 55 bolzano trentino italy alps landscape mountains cathedral

57 Purpose-based categorization of social tags Let s tag this Flickrphoto! 56 bolzano trentino italy alps landscape mountains cathedral italia alpi montagna duomo

58 Purpose-based categorization of social tags Let s tag this Flickrphoto! 57 bolzano trentino italy alps landscape mountains cathedral italia alpi montagna duomo beautiful unforgattablepictures holidays summer 2010 nikon 50mm

59 Purpose-based categorization of social tags 58 Users have different intentions when tagging, and not all tags available in a folksonomyare related to the content of the annotated items

60 Purpose-based categorization of social tags 59 Tagging purposes/intentions Cantador et al. Xu et al. Sen et al. Golder et al. Bischoff et al. Content-based What or who is about Topic Content-based Attribute Factual What it is Who owns it Type Author/owner Context-based Context-based Refining other categories Time Location Subjective Subjective Subjective Qualities/characteristics Opinions/qualities Organisational Organisational Personal Task organisation Self reference Usage context Self reference Cantador, I., Konstas, I., Jose, J. M CategorisingSocial Tags to Improve Folksonomy-based Recommendations. Journal of Web Semantics. Elsevier. In press. Xu, Z., Fu, Y., Mao, J., Su, D TowardstheSemanticWeb: CollaborativeTagSuggestions. In Proceedingsof thewww 06 Collaborative Web Tagging Workshop. Sen, S., Lam, S. K., Rashid, A. M., Cosley, D., Frankowski, D., Osterhouse, J., Harper, M. F., Riedl, J Tagging, Communities, Vocabulary, Evolution. In Proceedings of the 20th ACM Conference on Computer Supported Cooperative Work(CSCW 06), Golder, S. A., Huberman, B. A Usage Patterns of Collaborative Tagging Systems. Journal of Information Science 32(2), Bischoff, K., Firan, C. S., Nejdl, W., Paiu, R Can AllTagsBe UsedforSearch? In Proceedingof the17th ACM Conferenceon Information and Knowledge Management (CIKM 08),

61 Purpose-based categorization of social tags 60 4 categories Content-based tags - Describe the content of the items; e.g. vehicle, dog, tree Context-based tags - Provide contextual information; e.g. bolzano, mountain, summer, holidays Subjective tags - Express opinions and qualities; e.g. happy, sunny, contemporary art Organizational tags - Define personal usages and tasks; e.g. to look at, scan for print, myself, our best friends

62 Purpose-based categorization of social tags A couple of examples glasgow context kilt content 61 How do we infer the above categorization?

63 Purpose-based categorization of social tags 62 A couple of examples glasgow context kilt content How do we infer the above categorization? glasgow city in Scotland location context kilt Scotishpiece of cloth physical entity content

64 Purpose-based categorization of social tags 63 Research questions RQ1: How to automaticallycategorize tags based on their purpose/intention? RQ2: Is a purpose-oriented categorization of tags useful for folksonomy-based recommendationstrategies?

65 Purpose-based categorization of social tags 64 Research questions RQ1: How to automaticallycategorize tags based on their purpose/intention? RQ2: Is a purpose-oriented categorization of tags useful for folksonomy-based recommendation strategies? Yes

66 Purpose-based categorization of social tags 65 Proposed approach semantic expansion concept -category assignment tag-concept mapping 1a 2a 3a content-based & context-based tag categorization raw tag nyc [New_York_city] USA_cities > New_York_state_cities > Cities > Locations context to_read to <preposition> + read <verb> task organisational category natural 1b language processing 2b 3b subjective & organisational tag categorization part-of-speech analysis Categorization heuristic

67 Purpose-based categorization of social tags 66 Path A: tag-concept mapping Linked Data

68 Purpose-based categorization of social tags 67 Path A: tag-concept mapping DBPedia Giving structure to Wikipedia - 3.4M entities: 300K people, 400K places, 146K species, 140K organizations, 94K music albums, 49K films, 15K video games, K Wikipedia categories - 75K YAGO categories - 1.5M links to images - 5.5M links to web pages - 4.9M links to RDF repositories

69 Purpose-based categorization of social tags 68 Path A: tag-concept mapping YAGO (Yet Another Great Ontology) - DBPedia concepts + WordNet structure - >2M entities - Manually confirmed accuracy of 95% Suchanek, F. M., Kasneci, G., Weikum, G YAGO: A Core of Semantic Knowledge. In Proceedings of the 16th International Conference on World Wide Web (WWW'07), Suchanek, F. M., Kasneci, G., Weikum, G YAGO: A LargeOntologyfromWikipediaand WordNet. Journal of Web Semantics 6(3),

70 Purpose-based categorization of social tags 69 Path A: tag-concept mapping Exploiting YAGO structure: reference classes WordNet Wikipedia

71 Purpose-based categorization of social tags 70 Path A: tag-concept mapping Exploiting YAGO structure: reference classes

72 Purpose-based categorization of social tags 71 Path A: tag-concept mapping Tag-concept mapping examples (from a Flickr dataset)

73 Purpose-based categorization of social tags 72 Path A: tag-concept mapping Ambiguity java

74 Purpose-based categorization of social tags 73 Categorization results accuracy 30 subjects, 4K tag assignments

75 Purpose-based categorization of social tags 74 Open research lines need for improving disambiguation process definition of more POS-patterns revision of some POS-patterns: - e.g. subjective content-based tags [*<adjective><noun>*]: big house house

76 Contextualizationanddisambiguationoftags 75 What is the meaning of sf?

77 Contextualizationanddisambiguationoftags What is the meaning of sf? sf sanfrancisco california usa city bridge goldengate sea 76 sf sciencefiction fiction starwars movie tatooine androids c3po r2d2 Different meanings!

78 Contextualizationanddisambiguationoftags What is the meaning of web? 77 web webbrowser browser browsing opera explorer firefox chrome safari web socialweb web20 socialmedia facebook twitter flickr lastfm Different semantic contexts!

79 Contextualizationanddisambiguationoftags 78 Research questions RQ1: How to accurately and efficientlyidentify the actual meaning or semantic context of a tag? RQ2: Is a semantic contextualization of tags useful for folksonomy-based user modelling and recommendation strategies?

80 Contextualizationanddisambiguationoftags 79 Research questions RQ1: How to accurately and efficientlyidentify the actual meaning or semantic context of a tag? RQ2: Is a semantic contextualization of tags useful for folksonomy-based user modelling and recommendation strategies? Yes, again!

81 Contextualizationanddisambiguationoftags Proposed approach 80 Cantador, I., Bellogín, A., Fernández-Tobías, I., López-Hernández, S Semantic Contextualization of Social Tag-based Item Recommendations. In Proceedings of the 12th International Conference on E-Commerce and Web Technologies (EC-Web 2011), pp

82 Contextualizationanddisambiguationoftags 81 Aggregation methods ([user, item, tag] [item, tag]) Projection Distributional Macro-aggregation Collaborative Similarity measures ([item, tag] [tag, tag]) Matching Overlapping Jaccard Dice Cosine Mutual information Markines, B., Cattuto, C., Menczer, F., Benz, D., Hotho, A., Stumme, G Evaluating Similarity Measures for Emergent Semantics of Social tagging. In Proceedings of the 18th International Conference on World Wide Web (WWW'09),

83 Contextualizationanddisambiguationoftags Aggregation methods ([user, item, tag] [item, tag]) 82 Similarity measures ([item, tag] [tag, tag])

84 Contextualizationanddisambiguationoftags 83 Compute the semantic distance between each pair of tags Create the whole tag graphgbased on the above distances For each tag t, extract a subgraphg t from Gwith the tags most similar to t ClusteringG t by using the algorithm proposed by Newman and Girvan, 2004 Fast computation Automatic clustering stop criterion Newman, M. E. J., and Girvan, M Finding and Evaluating Community Structure in Networks. Physical Review, E 69, Newman, M. E. J Finding Community Structure in Networks Using the Eigenvectors of Matrices. Physical Review, E 74,

85 Contextualizationanddisambiguationoftags sfin Delicious 84

86 Contextualizationanddisambiguationoftags webin Delicious 85

87 Contextualizationanddisambiguationoftags Profile models Basic tag weight java 0.29 j2ee 0.17 web 0.42 tag 0.12 Tag:java Single-context tag tag context weight java programming 0.29 j2ee java 0.17 web socialweb 0.42 tag socialweb 0.12 Contextualized tag:java programming Extended single-context tag weight java 0.27 programming 0.05 j2ee 0.15 web 0.34 socialweb 0.09 tag 0.10 Multiple-context Extended multiple-context tag tag context weight tag weight java programming 0.20 java 0.30 java tools 0.14 programming 0.03 j2ee java 0.13 tools 0.02 web socialweb 0.23 j2ee 0.11 web webdev 0.21 web 0.37 tag socialweb 0.08 socialweb 0.05 webdev 0.04 tag Cantador, I., Bellogín, B., Vallet, D Content-based Recommendation in Social Tagging Systems. In Proceedings of the 4th ACM Conference on Recommender Systems (RecSys'10),

88 Contextualizationanddisambiguationoftags 87 Still evaluating the recommendation models by using semantically contextualized tag-based profiles... Some (preliminary) results with a Delicious dataset Improvements!

91 Recommendation of news items 90 Personalised and group-oriented content retrieval Multi-criteria evaluations Collaborative content-based recommendations Semantic search Automatic semantic annotation Long-term (profile) and short-term (context) user preferences Tags, comments and ratings Cantador, I., Bellogín., A., Castells, P A Semantic Web Approach to Recommending News. In Proc. of the 5th Intl. Conference on Adaptive Hypermedia and Adaptive Web-Based Systems (AH 2008),

92 Architecture 91 News services Dynamic web interface News Semantic Annotator RSS feeds Domain ontologies User preferences User tags User profile editor User preference learner User actions Annotated news items Ontology-based user profiles Personalised Context-aware Ontology-based recommendation algorithms Group-oriented Multi-facet

93 User profile editor 92 Semantic user preferences Ontology class browser Ontology instance viewer

94 93 Problem 1: the ontologies did not have instances - named entities: people, places, organizations, etc. e.g. teide Teide should be an instance of the class Volcano (subclass of Mountain) Problem 2: the news items did not have semantic annotations Problem 3: the users had to manually search for and select ontology concepts for building their profiles

95 Identification of ontology concepts for tags 94 Research questions RQ1: How to automatically populate ontologiesand annotate news items? RQ2: How to facilitate users to build their ontology-based profiles without asking them to search for concepts of interests? Approach Mapping terms and social tags to ontology concepts Cantador, I., Castells, P., Bellogín, A An Enhanced Semantic Layer for Hybrid Recommender Systems: Application to News Recommendation. International Journal on Semantic Web and Information Systems 7(1), pp Fernández, M., Cantador, I., López, V., Vallet, D., Castells, P., Motta, E Semantically Enhanced Information Retrieval: An Ontology-based Approach. Journal of Web Semantics 9(4), pp

96 Identification of ontology concepts for tags Proposal Users 95 Social network Tags Folksonomy Concepts Ontology Terms, tags Resources Content graph

97 Identification of ontology concepts for tags 96 Mapping a term or tag to a Wikipedia entry (+/- dealing with ambiguity) teide Information extraction and processing of Wikipedia categories teide volcanoes, mountains, spain,... Mapping of the term categories to ontology classes teide instance of volcano class

98 Identification of ontology concepts for tags 97 Ontology population 17 ontologies(built from IPTC taxonomy) 744 classes; 121,135 instances 69.9% / 84.4% of accuracy (20 people evaluating 10,200 instances)

99 Identification of ontology concepts for tags 98 Semantic annotation of news items 9,698 news items (gathered during two months) 66,378 generated annotations 7 annotations per item (average) 74.8% of accuracy (20 people evaluating 1,600 items)

100 Identification of ontology concepts for tags 99 (Some) Recommendation algorithms Content-based recommender Context-aware recommender Multi-layer hybrid recommender Semantic Knowledge Base Weighted Semantic Annotations Weighted Semantic Preferences Personalised Ranking Semantic User Profile Final Ranked Item List Item Repository Item Retrieving Unranked Items Cantador, I., Castells, P., Bellogín, A An Enhanced Semantic Layer for Hybrid Recommender Systems: Application to News Recommendation. International Journal on Semantic Web and Information Systems 7(1), pp

101 Identification of ontology concepts for tags 100 (Some) Recommendation algorithms Content-based recommender Context-aware recommender Multi-layer hybrid recommender Initial User Preferences Extended User Preferences Initial Runtime Context Extended Runtime Context Contextualized User Preferences Cantador, I., Castells, P., Bellogín, A An Enhanced Semantic Layer for Hybrid Recommender Systems: Application to News Recommendation. International Journal on Semantic Web and Information Systems 7(1), pp

102 Identification of ontology concepts for tags 101 (Some) Recommendation algorithms Content-based recommender Context-aware recommender Multi-layer hybrid recommender CoI obtained from cluster 1 CoI obtained from cluster 2 CoI obtained from cluster 3 CoI obtained from cluster 4 Cantador, I., Castells, P., Bellogín, A An Enhanced Semantic Layer for Hybrid Recommender Systems: Application to News Recommendation. International Journal on Semantic Web and Information Systems 7(1), pp

103 Identification of ontology concepts for tags 102 (Many) Recommendation results... A summary: Good results - Semantic expansion of ontology-based user preferences - Combination of personalized and contextualized approaches - Hybrid approach in cold-start and high-sparsity situations Open issues - Improving disambiguation - Addressing scalability - User preference learning (e.g.synonym) and recommendation

105 Cross-domain Recommendations Inter-domain semantic relatedness 104 Kaminskas, M., Fernández-Tobías, I., Ricci, F., Cantador, I Knowledge-based Identification of Music Suited for Places of Interest. Journal of Information Technology and Tourism 14(1), pp