GeniUS: Generic User Modeling Library
for the Social Semantic Web
JIST2011, December 2011, Hangzhou, China



                                     QiGao, Fabian Abel, Geert-Jan Houben
                                  {q.gao, f.abel, g.j.p.m.houben}@tudelft.nl
                                                   Web Information Systems
                                               Delft University of Technology



        Delft
        University of
        Technology
What we do: Science and Engineering for the
        Personal Web
domains: news social mediacultural heritage public datae-learning

         Personalized              Personalized
                                                               Adaptive Systems
       Recommendations               Search


                                 Analysis and
                                User Modeling


                          Semantic Enrichment,
                          Linkage and Alignment

                                                user/usage data


                                Social Web
                       GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   2
Motivation
                                      • Sparsity problem
             Product                     • do not have enough useful
          Recommender                      information for a (new) user


         ?         ?                  • Possible solution: gatheringuser data from
                                        other sources
                                         • But not all data may be relevant for
          User Modeling
                                           the given application context.
                                         • how to filter out user data that does
                                           not fit the target application context?
I’m a new user.
Recommend me
 some product



                          GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   3
Research Challenges of GeniUS
 Various applications in different domains
   Product        Movie           Hotel                         Product
recommender   recommender     recommender                    recommender


                 Profile

                   ?
                                                                        customized user
                                                                       profile construction


               Analysis and                               interested in:
              User Modeling
                                                    Product Movie location
          Semantic Enrichment
                                        How can we build a flexible and extensible
                                        user modeling functionality that adapts to
                                       the demands of a given application context?


                            GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   4
What is GeniUS?

• GeniUSis a topic and user modeling software library that
   • produces semantically meaningful profiles to enhance the
     interoperability of profiles between applications;

   • provides functionality for aggregating relevant information about a
     user from the Social Web;
   • generates domain-specific user profiles according to the information
     needs of different applications;

   • is flexible and extensible to serve different applications.


                     GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   5
GeniUS: Generic Topic and User Modeling Library
    for the Social Semantic Web
                                                Semantic Web




                                                        semantic data
                          Filter


                                                            enriched
             user data                                          items                user profiles         RDF
                          Item         items    Enrichmen               Weighting
                                                                                                        Serializatio
                         Fetcher                     t                  Function    interested in:
                                                                                                             n
                                                                                     product location
Social Web

                                                 Modeling                                                   RDF
                                               Configuration                                             Repository



                                   GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web           6
Item
                                                                                                       Fetcher
   GeniUS modules: Item Fetcher and Semantic Enrichment
                                                                                                      Enrichmen
                                                                                                           t




             raw content
                                           a <sioc:Post> ;                       sioc:has_topicdbpedia:Apple_Inc;
                           Twitter                                 SpotLight,
                                        dcterms:created … ;                      sioc:has_topicdbpedia:GarageBand;
                            API                                     Zemanta,
                                        sioc:has_creator …;                      sioc:has_topicdbpedia:Ipad;
                                                                   OpenCalais
                                        sioc:content … .

Social Web
                                                                                    Awesome, love the new
                                      Awesome, love the new                         Garageband for iPad #apple
                                                                                    GaragebandiPad#apple
                                      Garageband for iPad #apple


                                                                   dbpedia:GarageBand dbpedia:Ipad dbpedia:Apple_Inc




                                     GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web    7
Weighting
   GeniUS modules: Weighting Function and                                              Function
   RDF Serialization                                                                      RDF
                                                                                       Serializatio
                                                                                            n



                         weight(dbpedia:Garag
                         eBand)

                               weight(dbpedi
                               a:Jazz)
                 weight(dbpedia:Secon
     TF          d_Life)
                                                    RDF
   TF-IDF
                                                Serialization
Time-sensitive




                                                the weighted
                                                interests vocabulary




                                GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web    8
Filter
      GeniUS modules: Configuration and Filter                                      Modeling
                                                                                  Configuration



(Jazz, 0.5889)
                                                                                    (Second_Life,
                                                                                         0.4101)
(Second_Life,
     0.3114)           SELECT DISTINCT ?t WHERE {
                                        Filter
                         ? <rdf:type><dbpedia-owl:Software> }                       (GarageBand,
(GarageBand,                                                                             0.2158)
     0.1638)
                                                    enriched
                               items                    items
                   Twitter
                                        SpotLight               TF
                    API




                                         Modeling
                                       Configuration

                       GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web    9
GeniUS: Generic Topic and User Modeling Library
for the Social Semantic Web


Semantic Web




                GeniUS                     User Profile                Applications
                                              interested in:

                                              product location …



 Social Web



      How do user profiles generated by GeniUS support
              different types of applications?

                    GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   10
Analysis of Domain-specific User Profile Construction

• Dataset
   • 72 Twitter users (CS researchers) observed over a period of 6 months
     (>40,000 tweets)
   • a variety of topics mentioned in the tweets



• Research questions
   • 1. What are the characteristics of (complete) Twitter-based profiles
     generated with GeniUS ?

   • 2. Can domain-specific profiles be derived from Twitter activities ?

   • 3. What are the characteristics of such domain-specific profiles?

                      GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   11
Analysis of Domain-specific User Profile Construction
average number of entities: 1097.1




                                   average number of types: 35.0




    a potential to generate domain-specific profiles
    by categorizing entities according to their types
                 GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   12
Analysis of Domain-specific User Profile Construction
                  domain: location      the more specific the domain
 generic
                  domain: entertainment the smaller the profiles
 (all domains)
                 × domain: product




       Are the domain-specific user profiles beneficial for
      supporting different recommendation applications?

                       GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   13
Evaluation of Domain-specific User Profile Construction
• Task: Recommending domain-specific tweets
• Domains:
   • three domains: location, entertainment, product
   • three sub-domains of product: book, software, music
• Recommender algorithm: cosine similarity between profile
  and candidate item
• Ground truth: relevant (re-)tweets of users
• Candidate items: all the tweets posted during evaluation
  period                                 Recommendations = ?

           P(u)= ?         user profile


                                                                    time
                                                     1 month
                     GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   14
Evaluation results

    the domain-specific user modeling strategies
   improve the performance of recommendations




                                                     three different domains




               GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   15
Evaluation results
 The sub-domain-specific user modeling strategy
also improve the performance of recommendation.




                                                      three sub-domains
                                                          of product




 The user modeling quality varies only slightly
        between the different domains
                 GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   16
Wrap up
• GeniUS: Generic topic and User modeling library for the Social Semantic Web
   • exploits traces (e.g. tweets) that people leave on the Social Web
   • enriches the semantics of these traces
   • constructs semantic user profiles
    profile construction can be customized and is adapted to a given application context

• Analysis:
   • Twitter-based user profiles contain a great variety of topics
   • GeniUS succeeds in generating profiles for different applications and domains

• Evaluation:
   • domain-specific user modeling strategies (powered by the semantic filtering of
     GeniUS) allow clearly for the best performance
   • the more GeniUS adapts to the given domain (and application context) the better
     the performance

                           GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   17
Thank You!


                     q.gao@tudelft.nl
                     Twitter: @qigaosh
QiGao
                     http://wis.ewi.tudelft.nl/tweetum/
                     http://wis.ewi.tudelft.nl/genius/
  GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web   18

GeniUS: Generic User Modeling Library for the Social Semantic Web

  • 1.
    GeniUS: Generic UserModeling Library for the Social Semantic Web JIST2011, December 2011, Hangzhou, China QiGao, Fabian Abel, Geert-Jan Houben {q.gao, f.abel, g.j.p.m.houben}@tudelft.nl Web Information Systems Delft University of Technology Delft University of Technology
  • 2.
    What we do:Science and Engineering for the Personal Web domains: news social mediacultural heritage public datae-learning Personalized Personalized Adaptive Systems Recommendations Search Analysis and User Modeling Semantic Enrichment, Linkage and Alignment user/usage data Social Web GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 2
  • 3.
    Motivation • Sparsity problem Product • do not have enough useful Recommender information for a (new) user ? ? • Possible solution: gatheringuser data from other sources • But not all data may be relevant for User Modeling the given application context. • how to filter out user data that does not fit the target application context? I’m a new user. Recommend me some product GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 3
  • 4.
    Research Challenges ofGeniUS Various applications in different domains Product Movie Hotel Product recommender recommender recommender recommender Profile ? customized user profile construction Analysis and interested in: User Modeling Product Movie location Semantic Enrichment How can we build a flexible and extensible user modeling functionality that adapts to the demands of a given application context? GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 4
  • 5.
    What is GeniUS? •GeniUSis a topic and user modeling software library that • produces semantically meaningful profiles to enhance the interoperability of profiles between applications; • provides functionality for aggregating relevant information about a user from the Social Web; • generates domain-specific user profiles according to the information needs of different applications; • is flexible and extensible to serve different applications. GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 5
  • 6.
    GeniUS: Generic Topicand User Modeling Library for the Social Semantic Web Semantic Web semantic data Filter enriched user data items user profiles RDF Item items Enrichmen Weighting Serializatio Fetcher t Function interested in: n product location Social Web Modeling RDF Configuration Repository GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 6
  • 7.
    Item Fetcher GeniUS modules: Item Fetcher and Semantic Enrichment Enrichmen t raw content a <sioc:Post> ; sioc:has_topicdbpedia:Apple_Inc; Twitter SpotLight, dcterms:created … ; sioc:has_topicdbpedia:GarageBand; API Zemanta, sioc:has_creator …; sioc:has_topicdbpedia:Ipad; OpenCalais sioc:content … . Social Web Awesome, love the new Awesome, love the new Garageband for iPad #apple GaragebandiPad#apple Garageband for iPad #apple dbpedia:GarageBand dbpedia:Ipad dbpedia:Apple_Inc GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 7
  • 8.
    Weighting GeniUS modules: Weighting Function and Function RDF Serialization RDF Serializatio n weight(dbpedia:Garag eBand) weight(dbpedi a:Jazz) weight(dbpedia:Secon TF d_Life) RDF TF-IDF Serialization Time-sensitive the weighted interests vocabulary GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 8
  • 9.
    Filter GeniUS modules: Configuration and Filter Modeling Configuration (Jazz, 0.5889) (Second_Life, 0.4101) (Second_Life, 0.3114) SELECT DISTINCT ?t WHERE { Filter ? <rdf:type><dbpedia-owl:Software> } (GarageBand, (GarageBand, 0.2158) 0.1638) enriched items items Twitter SpotLight TF API Modeling Configuration GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 9
  • 10.
    GeniUS: Generic Topicand User Modeling Library for the Social Semantic Web Semantic Web GeniUS User Profile Applications interested in: product location … Social Web How do user profiles generated by GeniUS support different types of applications? GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 10
  • 11.
    Analysis of Domain-specificUser Profile Construction • Dataset • 72 Twitter users (CS researchers) observed over a period of 6 months (>40,000 tweets) • a variety of topics mentioned in the tweets • Research questions • 1. What are the characteristics of (complete) Twitter-based profiles generated with GeniUS ? • 2. Can domain-specific profiles be derived from Twitter activities ? • 3. What are the characteristics of such domain-specific profiles? GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 11
  • 12.
    Analysis of Domain-specificUser Profile Construction average number of entities: 1097.1 average number of types: 35.0 a potential to generate domain-specific profiles by categorizing entities according to their types GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 12
  • 13.
    Analysis of Domain-specificUser Profile Construction domain: location the more specific the domain generic domain: entertainment the smaller the profiles (all domains) × domain: product Are the domain-specific user profiles beneficial for supporting different recommendation applications? GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 13
  • 14.
    Evaluation of Domain-specificUser Profile Construction • Task: Recommending domain-specific tweets • Domains: • three domains: location, entertainment, product • three sub-domains of product: book, software, music • Recommender algorithm: cosine similarity between profile and candidate item • Ground truth: relevant (re-)tweets of users • Candidate items: all the tweets posted during evaluation period Recommendations = ? P(u)= ? user profile time 1 month GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 14
  • 15.
    Evaluation results the domain-specific user modeling strategies improve the performance of recommendations three different domains GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 15
  • 16.
    Evaluation results Thesub-domain-specific user modeling strategy also improve the performance of recommendation. three sub-domains of product The user modeling quality varies only slightly between the different domains GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 16
  • 17.
    Wrap up • GeniUS:Generic topic and User modeling library for the Social Semantic Web • exploits traces (e.g. tweets) that people leave on the Social Web • enriches the semantics of these traces • constructs semantic user profiles  profile construction can be customized and is adapted to a given application context • Analysis: • Twitter-based user profiles contain a great variety of topics • GeniUS succeeds in generating profiles for different applications and domains • Evaluation: • domain-specific user modeling strategies (powered by the semantic filtering of GeniUS) allow clearly for the best performance • the more GeniUS adapts to the given domain (and application context) the better the performance GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 17
  • 18.
    Thank You! q.gao@tudelft.nl Twitter: @qigaosh QiGao http://wis.ewi.tudelft.nl/tweetum/ http://wis.ewi.tudelft.nl/genius/ GeniUS: Generic User ModelingLibraryfor the SocialSemantic Web 18

Editor's Notes

  • #3 Redundant;
  • #4 Spars
  • #5 Research challenges for GeniUSUser profiles with
  • #8 Item fetcher , enrichment
  • #9 Weighting function/RDF representationTime-sensitive, emphasize the temporal factor in news recommendations.
  • #10 Modeling configuration: specify the implementation of the different modules. Filtering function
  • #12 more details on the dataset
  • #14 Our hypothesis is that
  • #18 The user modeling quality varies only slightly between the different domains