Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
SASWeb 2012 - Workshop on Social, Semantic and Adaptive Web            Montréal (Canada), 16.07.2012      ‘Meaning is its ...
“meaning                                                                                      is its use”                 ...
semantics:                                             study of meaning                                                   ...
semantics plays a key role for                                   most of the adaptive systems.Cataldo Musto, ‘Meaning is i...
adaptive systems can benefit      from semantic representation              of the information.Cataldo Musto, ‘Meaning is i...
example.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender sys...
Recommender Systems  Relevant items (movies, news, books, etc.) are pushed to the  user according to her preferences or he...
content-based recommenders         Suggest items similar to those liked in the past by the userCataldo Musto, ‘Meaning is ...
scenario.                                            book recommendationCataldo Musto, ‘Meaning is its use’: towards the u...
content-based recommenders                             book recommendation: key concepts       • Each book has to be descr...
“I love turkey.          I will choose it for my holidays”                                                                ...
“what book will the                                  user be interested in?”                                              ...
how can we boost                          content-based recommender systems                       with semantics ?Cataldo ...
alternative representation         distributional models                                                           (Firth,...
distributional models  insight by analyzing large corpus of textual data it is possible to infer information about the usa...
distributional models  insight by analyzing large corpus of textual data it is possible to infer information about the usa...
distributional models  insight by analyzing large corpus of textual data it is possible to infer information about the usa...
distributional models   •      Key: definition of what is the          ‘context’         • Different granularities         ...
distributional models                                 term/context matrix (WordSpace)                                 c1  ...
WordSpace                                                                    example                                      ...
distributional models                                         beer vs. glass: good overlap                                ...
distributional models                                           beer vs. spoon: no overlap                                ...
representation of documents can be inferred     by combining the representation of the terms               occurring in th...
distributional models                                   term/context matrix (DocSpace)                                    ...
distributional models                       similarity between documents (DocSpace)                                       ...
distributional models                    similarity between documents: good overlap                                       ...
distributional models                       similarity between documents: no overlap                                      ...
distributional models                                                                 recap     models for representing te...
strength:            representation based on         distributional models are             inherently multilingual.Cataldo...
distributional models                                            multilingual representation                       • Assum...
distributional models                                            multilingual representation                        • Assu...
(english) WordSpace                                                            beer                                       ...
(italian) WordSpace                                                                                            relationshi...
multilingual representation                     comes with no costs.                      Thanks to distributional hypothe...
distributional models                                                                 recap     models for representing te...
how to combine             distributional models                                                                          ...
a novel recommendation framework based on VSM                                                          eVSM               ...
eVSM building blocks               distributional models.Cataldo Musto, ‘Meaning is its use’: towards the use of distribut...
eVSM representation                                                   mystery book                                        ...
user profile                                                  how to represent it?    • In eVSM each item is represented as...
user profile                                        Items                                Rating                        Thre...
eVSM scenario                                                   mystery book                                              ...
eVSM scenario                                                                                                          Rec...
eVSM scenario                                                                                                           re...
Why a multilanguage        representation does matter?Cataldo Musto, ‘Meaning is its use’: towards the use of distribution...
Language issues              • VSM representation is language-dependant               • User profile built in a language ca...
eVSM                               language-dependant recommendations                            user profile              ...
Vector Space Model                                                       multilingual scenario                            ...
Vector Space Model                                                       multilingual scenario                            ...
Vector Space Model                                                       multilingual scenario                            ...
eVSM                                                       multilingual scenario                                          ...
eVSM                               language-dependant recommendations                                                     ...
eVSM                               language-dependant recommendations                                                     ...
eVSM overcomes this issue.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-bas...
eVSM                          providing suggestions - multilingual scenario      DocSpace for L1                          ...
eVSM                          providing suggestions - multilingual scenario      DocSpace for L1                          ...
eVSM                          providing suggestions - multilingual scenario      DocSpace for L1                          ...
eVSM                          providing suggestions - multilingual scenario      DocSpace for L1                          ...
eVSM                          providing suggestions - multilingual scenario      DocSpace for L1                          ...
eVSM                          providing suggestions - multilingual scenario      DocSpace for L1                          ...
eVSM                                       multilanguage recommendations                                                  ...
recap and contributions.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based...
adaptive systems.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recomm...
semantics.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender s...
recommender systems.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based rec...
eVSMCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems...
richer representation based on                  distributional modelsCataldo Musto, ‘Meaning is its use’: towards the use ...
framework for multilingual recommendationsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics...
experimental evaluation                                       applicationsCataldo Musto, ‘Meaning is its use’: towards the...
applications                                                      ‘in vitro’                                              ...
movie recommendation                                                   ‘in vitro’ experiments   • Goal: to provide users w...
experimental design                                                              experiments   • Experiment    • How do th...
experiment             87                                                     Movielens dataset                           ...
experiment             87                                                     Movielens dataset                           ...
applications                                                        ‘in vivo’                                             ...
Play.me                                           personalized music playlists          •     Goal               •     To ...
Play.me                                                              architectureCataldo Musto, ‘Meaning is its use’: towa...
Play.me                                                              architectureCataldo Musto, ‘Meaning is its use’: towa...
Extractor                                                                      insight Social Media provide us unlimited, ...
Extractor                                                                      insight        Social Media are a cheap and...
Myusic                                      data extraction from FacebookCataldo Musto, ‘Meaning is its use’: towards the ...
Myusic                                      data extraction from Facebook            explicit preferencesCataldo Musto, ‘M...
Myusic                                      data extraction from Facebook            implicit preferencesCataldo Musto, ‘M...
Play.me                                                              architectureCataldo Musto, ‘Meaning is its use’: towa...
Play.fm                                                               enrichment     • Given a set of explicit preferences...
Play.fm                                                   enrichment example                                              ...
Play.fm                                                               enrichment     • Comparison of two approaches       ...
Linked Open Data Cloud                                                                                            Structur...
Play.fm                       enrichment based on Distributional Models           • Distributional Models           • Each...
Play.fm                       enrichment based on Distributional Models                                                   ...
Play.fm                       enrichment based on Distributional Models                              input: vector space r...
Play.me                                                              architectureCataldo Musto, ‘Meaning is its use’: towa...
Play.me                                                                     playlistMost popular songs of the artists extr...
experimental design    • Experiment     • Which one is the enrichment technique that                can provide users with...
experimental design                                                                     results                          7...
conclusions.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender...
eVSM overcomes                                      state of the art approachesCataldo Musto, ‘Meaning is its use’: toward...
semantic representation based                       on distributional models                 effectively tackles language-...
end.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems...
questions?Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender s...
Upcoming SlideShare
Loading in …5
×

'Meaning is its use' - Towards the use of distributional semantics for content-based recommender systems

999 views

Published on

Invited talk at UMAP 2012 Conference, Workshop on Social Adaptive and Semantic Web

Published in: Technology, Spiritual
  • Be the first to comment

'Meaning is its use' - Towards the use of distributional semantics for content-based recommender systems

  1. 1. SASWeb 2012 - Workshop on Social, Semantic and Adaptive Web Montréal (Canada), 16.07.2012 ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems Cataldo Musto, Ph.D. University of Bari Aldo Moro (Italy) - cataldo.musto@uniba.it
  2. 2. “meaning is its use” L.Wittgenstein (Austrian philosopher)Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  3. 3. semantics: study of meaning Greek: σημαντικοςCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  4. 4. semantics plays a key role for most of the adaptive systems.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  5. 5. adaptive systems can benefit from semantic representation of the information.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  6. 6. example.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  7. 7. Recommender Systems Relevant items (movies, news, books, etc.) are pushed to the user according to her preferences or her needs.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  8. 8. content-based recommenders Suggest items similar to those liked in the past by the userCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  9. 9. scenario. book recommendationCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  10. 10. content-based recommenders book recommendation: key concepts • Each book has to be described through a set of textual features • e.g title of the book, summary, etc. • Each user is described through textual features, as well • Recommendations are provided by calculating the overlap between the textual description of the TV show and the features stored in the user profileCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  11. 11. “I love turkey. I will choose it for my holidays” ?Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  12. 12. “what book will the user be interested in?” vs.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  13. 13. how can we boost content-based recommender systems with semantics ?Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  14. 14. alternative representation distributional models (Firth, 1957) Firth, J.R. A synopsis of linguistic theory 1930-1955. In Studies in Linguistic Analysis, pp. 1-32, 1957.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  15. 15. distributional models insight by analyzing large corpus of textual data it is possible to infer information about the usage (about the meaning) of the terms.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  16. 16. distributional models insight by analyzing large corpus of textual data it is possible to infer information about the usage (about the meaning) of the terms. exampleCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  17. 17. distributional models insight by analyzing large corpus of textual data it is possible to infer information about the usage (about the meaning) of the terms. distributional hypothesis “ words that share similar contexts (usages) share similar meaning “Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  18. 18. distributional models • Key: definition of what is the ‘context’ • Different granularities are possible • Document • Paragraph • Sentence • Sliding window of wordsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  19. 19. distributional models term/context matrix (WordSpace) c1 c2 c3 c4 c5 c6 c7 c8 c9 t1 ✔ ✔ ✔ ✔ t2 ✔ ✔ ✔ ✔ t3 ✔ ✔ ✔ t4 ✔ ✔ ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  20. 20. WordSpace example beer wine glass spoonCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  21. 21. distributional models beer vs. glass: good overlap c1 c2 c3 c4 c5 c6 c7 c8 c9 t1 ✔ ✔ ✔ ✔ t2 ✔ ✔ ✔ ✔ t3 ✔ ✔ ✔ t4 ✔ ✔ ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  22. 22. distributional models beer vs. spoon: no overlap c1 c2 c3 c4 c5 c6 c7 c8 c9 t1 ✔ ✔ ✔ ✔ t2 ✔ ✔ ✔ ✔ t3 ✔ ✔ ✔ t4 ✔ ✔ ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  23. 23. representation of documents can be inferred by combining the representation of the terms occurring in the document.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  24. 24. distributional models term/context matrix (DocSpace) c1 c2 c3 c4 c5 c6 c7 c8 c9 t2 ✔ ✔ ✔ ✔ t3 ✔ ✔ ✔ d1 ✔ ✔ ✔ ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  25. 25. distributional models similarity between documents (DocSpace) c1 c2 c3 c4 c5 c6 c7 c8 c9 d1 ✔ ✔ ✔ ✔ ✔ d2 ✔ ✔ ✔ d3 ✔ ✔ d4 ✔ ✔ ✔ ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  26. 26. distributional models similarity between documents: good overlap c1 c2 c3 c4 c5 c6 c7 c8 c9 d1 ✔ ✔ ✔ ✔ ✔ d2 ✔ ✔ ✔ d3 ✔ ✔ d4 ✔ ✔ ✔ ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  27. 27. distributional models similarity between documents: no overlap c1 c2 c3 c4 c5 c6 c7 c8 c9 d1 ✔ ✔ ✔ ✔ ✔ d2 ✔ ✔ ✔ d3 ✔ ✔ d4 ✔ ✔ ✔ ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  28. 28. distributional models recap models for representing terms/ documents in large vector spaces light semantics it is simple to calculate similarities between words and documentCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  29. 29. strength: representation based on distributional models are inherently multilingual.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  30. 30. distributional models multilingual representation • Assumption • The distribution of the terms is (almost) language- independent drink bere beer / birra glass bicchiereCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  31. 31. distributional models multilingual representation • Assumption • The distribution of the terms is (almost) language- independent The position of concept of in a WordSpace will be beer always the same, regardless the language!Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  32. 32. (english) WordSpace beer wine spoon dogCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  33. 33. (italian) WordSpace relationships between terms stay birra regardless the language! vino cucchiaio caneCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  34. 34. multilingual representation comes with no costs. Thanks to distributional hypothesis.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  35. 35. distributional models recap models for representing terms/ documents in large vector spaces light semantics it is simple to calculate similarities between words and document representation is inherently multilanguageCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  36. 36. how to combine distributional models with content-based recommender systems?Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  37. 37. a novel recommendation framework based on VSM eVSM enhanced Vector Space ModelCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  38. 38. eVSM building blocks distributional models.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  39. 39. eVSM representation mystery book poetry book recipe book recipe bookCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  40. 40. user profile how to represent it? • In eVSM each item is represented as a vector • User profile vector space representation as well needs a • How? • For example, by combining vectors of the items (documents) the user liked in the pastCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  41. 41. user profile Items Rating Threshold VSM representation of RI-based profile for user uCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  42. 42. eVSM scenario mystery book user profile poetry book recipe book recipe bookCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  43. 43. eVSM scenario Recommendation task seen as mystery book similarity user profile calculation between vectors poetry book recipe book recipe bookCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  44. 44. eVSM scenario recommender mystery book system suggests user profile mystery book. poetry book recipe book recipe bookCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  45. 45. Why a multilanguage representation does matter?Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  46. 46. Language issues • VSM representation is language-dependant • User profile built in a language can not be exploited to provide recommendation of items described in another languageCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  47. 47. eVSM language-dependant recommendations user profile content-based recommendationsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  48. 48. Vector Space Model multilingual scenario d1 d2 d3 p t1 basketball italian ✔ ✔ document t2 italian ✔ ✔ t3 bargnani ✔ ✔ t4 pallacanestro ✔ ✔ t5 italiana ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  49. 49. Vector Space Model multilingual scenario d1 d2 d3 p t1 basketball english ✔ ✔ t2 italian documents ✔ ✔ t3 bargnani ✔ ✔ t4 pallacanestro ✔ ✔ t5 italiana ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  50. 50. Vector Space Model multilingual scenario d1 d2 d3 p t1 basketball ✔ ✔ t2 italian ✔ ✔ t3 bargnani ✔ ✔ t4 pallacanestro ✔ ✔ t5 italiana ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  51. 51. eVSM multilingual scenario d1 d2 d3 p t1 basketball user interested in ✔ ✔ t2 italian basketball, italian ✔ ✔ language t3 bargnani ✔ ✔ t4 pallacanestro ✔ ✔ t5 italiana ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  52. 52. eVSM language-dependant recommendations d1 d2 d3 p t1 basketball ✔ ✔ t2 italian ✔ ✔ t3 bargnani ✔ ✔ t4 pallacanestro ✔ ✔ t5 italiana ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  53. 53. eVSM language-dependant recommendations d1 d2 d3 p t1 basketball ✔ ✔ X t2 italian ✔ ✔ X t3 bargnani ✔ ✔ t4 pallacanestro ✔ ✔ t5 italiana ✔ ✔Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  54. 54. eVSM overcomes this issue.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  55. 55. eVSM providing suggestions - multilingual scenario DocSpace for L1 DocSpace for L2 c1 c2 c3 c4 c5 . . . ck c1 c2 c3 c4 c5 . . . ck d1 Parallel d1 DocSpaces d2 d2 Built upon the d3 same d3 d4 set of content d4 d5 d5 (italian) (english)Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  56. 56. eVSM providing suggestions - multilingual scenario DocSpace for L1 DocSpace for L2 italian football news english football news italian user profile english basketball news italian basketball news english politics news italian politics news english politics italian politics news news (italian) (english)Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  57. 57. eVSM providing suggestions - multilingual scenario DocSpace for L1 DocSpace for L2 c1 c2 c3 c4 c5 . . . ck c1 c2 c3 c4 c5 . . . ck d1 Parallel d1 DocSpaces d2 d2 Built upon the d3 same d3 d4 set of content d4 p d5 L1 user profile in L1 (italian)Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  58. 58. eVSM providing suggestions - multilingual scenario DocSpace for L1 DocSpace for L2 c1 c2 c3 c4 c5 . . . ck c1 c2 c3 c4 c5 . . . ck d1 Parallel d1 DocSpaces d2 d2 Built upon the d3 same d3 d4 set of content d4 p p L1 L1 we can project user profile in the DocSpace of english itemsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  59. 59. eVSM providing suggestions - multilingual scenario DocSpace for L1 DocSpace for L2 italian football news english football news italian user profile english basketball news italian basketball news english politics news italian politics news english politics italian politics news news (italian) (english)Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  60. 60. eVSM providing suggestions - multilingual scenario DocSpace for L1 DocSpace for L2 italian football news english football news italian user profile italian user profile english basketball news italian basketball news english politics news italian politics news english politics italian politics news news (italian) (english) through similarity calculations an english news about basketball is received as recommendation!Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  61. 61. eVSM multilanguage recommendations italian englishCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  62. 62. recap and contributions.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  63. 63. adaptive systems.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  64. 64. semantics.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  65. 65. recommender systems.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  66. 66. eVSMCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  67. 67. richer representation based on distributional modelsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  68. 68. framework for multilingual recommendationsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  69. 69. experimental evaluation applicationsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  70. 70. applications ‘in vitro’ experimentsCataldo Musto - Enhanced Vector Space Models for Content-based Recommender Systems - Ph.D. defense - University of Bari Aldo Moro, Italy - 08.06.12
  71. 71. movie recommendation ‘in vitro’ experiments • Goal: to provide users with recommendations about movies worth to be watched. • Subset of 100k MovieLens dataset + Wikipedia content • Monolingual and Multilingual settingsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  72. 72. experimental design experiments • Experiment • How do the model perform with respect to other state of the art approaches? • VSM - Vector Space Model • LSI - Latent Semantic Indexing • Bayes Bayes Text ClassifierCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  73. 73. experiment 87 Movielens dataset eVSM VSM 86.25 85.94 86.01 LSI Bayes 85.58 85.52 85.5 85.39 85.27 84.97 84.85 84.77 84.75 84.75 84.7 84.7 84.58 84.47 84.5 84.43 84 p@1 P@3 P@5 P@10 Gap always around 1%Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  74. 74. experiment 87 Movielens dataset eVSM VSM 86.25 85.94 86.01 LSI Bayes 85.58 85.52 85.5 85.39 85.27 84.97 84.85 84.77 84.75 84.75 84.7 84.7 84.58 84.47 84.5 84.43 84 p@1 P@3 P@5 P@10 Significant ImprovementCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  75. 75. applications ‘in vivo’ experimentsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  76. 76. Play.me personalized music playlists • Goal • To provide users with personalized music playlists • Methodology • Extraction of explicit user preferences from Facebook • Playlist creation by enriching explicit user preferences with similar artists. • Comparison of two enrichment algorithms • DBPedia-based enrichment • Distributional models-based enrichmentCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  77. 77. Play.me architectureCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  78. 78. Play.me architectureCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  79. 79. Extractor insight Social Media provide us unlimited, trustful and continously updated flow of information about user interests and needs.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  80. 80. Extractor insight Social Media are a cheap and effective way to overcome cold start.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  81. 81. Myusic data extraction from FacebookCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  82. 82. Myusic data extraction from Facebook explicit preferencesCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  83. 83. Myusic data extraction from Facebook implicit preferencesCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  84. 84. Play.me architectureCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  85. 85. Play.fm enrichment • Given a set of explicit preferences extracted from Facebook • Play.me enrichs this set by calculating artists similar to those the user explicity likesCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  86. 86. Play.fm enrichment example Coldplay extracted from Facebook enrichment radiohead red hot chili peppers kings of leonCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  87. 87. Play.fm enrichment • Comparison of two approaches • Linked Data • Distributional ModelsCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  88. 88. Linked Open Data Cloud Structured (RDF) representation of the information stored in Wikipedia.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  89. 89. Play.fm enrichment based on Distributional Models • Distributional Models • Each artist is represented through a set of tags • Each artist is represented as a point in a distributional DocSpace • Similarity calculations to extract the most similar artists.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  90. 90. Play.fm enrichment based on Distributional Models Coldplay Radiohead Kings of Leon Lady GagaCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  91. 91. Play.fm enrichment based on Distributional Models input: vector space representation output: artists with the highest cosine similarity radiohead the killers kings of leonCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  92. 92. Play.me architectureCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  93. 93. Play.me playlistMost popular songs of the artists extracted from Facebook (as well as those added through the enrichment) are proposed to the user.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  94. 94. experimental design • Experiment • Which one is the enrichment technique that can provide users with the best playlists ? • User study with 30 users.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  95. 95. experimental design results 76.3 80 75.2 Linked Data Distributional Models Baseline (Popularity) 73.75 69.7 67.5 65.9 64.6 61.25 63.2 58 58 58 55 m=1 m=2 m=3 m = number of artists added for each extracted artistCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  96. 96. conclusions.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  97. 97. eVSM overcomes state of the art approachesCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  98. 98. semantic representation based on distributional models effectively tackles language-issues of CBRSCataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  99. 99. end.Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12
  100. 100. questions?Cataldo Musto, ‘Meaning is its use’: towards the use of distributional semantics for content-based recommender systems. SASWeb Workshop, UMAP 2012, 16.07.12

×