際際滷

際際滷Share a Scribd company logo

STR-MT-IJCAI2011

Marc Torrens
Marc Torrens

The document summarizes lessons learned from developing, deploying, and operating real-world recommender systems. It discusses 10 key lessons, including making sure a recommender is needed based on factors like catalog and customer diversity, ensuring the recommender makes strategic sense for the business goals, and choosing the right partner such as a vendor, students, or experienced company depending on business size and type.

1 of 91
IJCAI, Industry Day Barcelona :: July 22, 2011 Top 10 Lessons learned Developing, Deploying and Operating Real-World Recommender Systems Marc Torrens Chief Innovation Officer Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 2 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 3 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 4 Year 2003... Provide Recommendations in the Music Space implicit preferences! What to play?What to synch? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Strands develops technologies to better understand peoples taste and help them discover things they like and didnt know about. About Strands 5 2003 2004 2005 2006 2007 2008 2009 2010 music people videos music Strands Recommender Strands Fitness Strands Finance Same mission evolving through different domains 2011 RecSys06 summer school RecSys07 Minneapolis RecSys08 Lausanne RecSys09 NY RecSys10 Barcelona RecSys11 Chicago Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 6 Understanding consumer habits at the commerce (transaction) level Understanding consumer preferences in real-world activities Understanding consumer behavior at the product level Highly-targeted Product Placement Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 7 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 8 BBVA, Spain ING, Netherlands BMO, Canada Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 9 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 10 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 11 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Why Personalize? 12 The Paradox of Choice by Barry Schwartz Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser IMPORTANT UNCOMFORTABLE CHALLENGING OTHER POINTS OF VIEWS Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 14 A Recommender selects the product that if acquired by the buyer maximizes value of both buyer and seller at a given point in time Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Recommender Components Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Knowledge Base Recommender Components Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Knowledge Processing Application Knowledge Base Recommender Components Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 25%25%25%25% Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge User Interface It has a certain level of autonomy presenting recommendations to the end user Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 25%25%25%25% Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 16 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 1. Do I need a recommender? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 1. Do I need a recommender? 2. How do I get one? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 1. Do I need a recommender? 2. How do I get one? 3. Is it performing well? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 1. Do I need a recommender? 2. How do I get one? 3. Is it performing well? 4. Was it a good idea after all? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 18 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 19 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 20 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 21 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 22 X X X X X 000s in both products and customers low medium high lowmediumhigh DiversityoftheCatalog Diversity of the Customers OK Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI showing this impact is already challenging! random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 24 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 25 Is the best recommendation for the customer the best recommendation for the business? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 26 Relevant vs Useful Correctness is often too obvious to be useful Riskier recommendations have less chances of being known customer perspective business perspective Short-term gain vs long-term return Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % How much business logic goes into Recommender? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % How much business logic goes into Recommender? Whats the right level of autonomy a recommender must have? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % How much business logic goes into Recommender? Whats the right level of autonomy a recommender must have? How can the business control recommendations? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 28 1. Do I need a recommender? 2. How do I get one? 3. Is it performing well? 4. Was it a good idea after all? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 29 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Small company Select a vendor Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Small company Select a vendor Medium company Hire a copule of PhD students! Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Small company Select a vendor Medium company Hire a copule of PhD students! Large company Partner with an experienced vendor Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Small company Select a vendor Medium company Hire a copule of PhD students! Large company Partner with an experienced vendor Tech company Organize a contest! Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 31 2000 2002 2003 2004 2006 2007 2008 20092001 20102005 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 31 2000 2002 2003 2004 2006 2007 2008 20092001 20102005 000s licensing + 000s integration Price $99 / month Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 31 2000 2002 2003 2004 2006 2007 2008 20092001 20102005 Vendors hundreds vendors few vendors 000s licensing + 000s integration Price $99 / month Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 4 Cold start? Be creative! 32 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 4 Cold start? Be creative! 33 With the advent of the Internet the start for a Recommender isnt so cold anymore Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 5 Data and algorithms. 34 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 5 Data and Algorithms. 35 Which really makes the difference? Ingredients or Receipe? Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 5 Data and Algorithms. 36 X X bad good badgood DataQuantityandQuality Algorithm Performance OK Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 6 Finding correlated items is easy, deciding what, how, and when to present to the user is hard. 37 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 6 Finding correlated items is easy, deciding what, how, and when to present to the user is hard. 38 Math Art Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 7 Dont wast time calculating nearest neighbors. 39 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 7 Dont waste time calculating nearest neighbours. 40 Let people tell you! Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 8 Dont wait too long to get ready to scale. 41 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 8 Dont wait too long to get ready to scale. 42 When is the right moment? if you do too soon and recommendations dont take off... if you do too late and recommendations do take off... Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 43 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 44 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 45 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 46 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 47 Implicit Ratings vs Explicit Ratings Implicit Semantics vs Explicit Semantic of Ratings Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 48 The Ideal (explicit) Rating system... feedback Good Badso-so* *optional at it may help to confirm some implicit actions. actions I have it (i knew it, i saw it) Dont show it any more Show it to me later has it (knew it, saw it)my friend Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 49 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 10 Measure everything. 50 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 10 Measure everything. 51 4th ACM Conference on Recommender Systems Barcelona :: September 30, 2010 Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 52 LESSON 10 Measure everything. Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 50% Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 50%20% Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 50%20%5% Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 50%20%5%25% Sunday, July 17, 2011
IJCAI, Industry Day Barcelona :: July 22, 2011 Questions? 54 Thank you! Sunday, July 17, 2011

More Related Content

STR-MT-IJCAI2011

  • 1. IJCAI, Industry Day Barcelona :: July 22, 2011 Top 10 Lessons learned Developing, Deploying and Operating Real-World Recommender Systems Marc Torrens Chief Innovation Officer Sunday, July 17, 2011
  • 2. IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 2 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
  • 3. IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 3 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
  • 4. IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 4 Year 2003... Provide Recommendations in the Music Space implicit preferences! What to play?What to synch? Sunday, July 17, 2011
  • 5. IJCAI, Industry Day Barcelona :: July 22, 2011 Strands develops technologies to better understand peoples taste and help them discover things they like and didnt know about. About Strands 5 2003 2004 2005 2006 2007 2008 2009 2010 music people videos music Strands Recommender Strands Fitness Strands Finance Same mission evolving through different domains 2011 RecSys06 summer school RecSys07 Minneapolis RecSys08 Lausanne RecSys09 NY RecSys10 Barcelona RecSys11 Chicago Sunday, July 17, 2011
  • 6. IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 6 Understanding consumer habits at the commerce (transaction) level Understanding consumer preferences in real-world activities Understanding consumer behavior at the product level Highly-targeted Product Placement Sunday, July 17, 2011
  • 7. IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 7 Sunday, July 17, 2011
  • 8. IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 8 BBVA, Spain ING, Netherlands BMO, Canada Sunday, July 17, 2011
  • 9. IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 9 Sunday, July 17, 2011
  • 10. IJCAI, Industry Day Barcelona :: July 22, 2011 About Strands 10 Sunday, July 17, 2011
  • 11. IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 11 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
  • 12. IJCAI, Industry Day Barcelona :: July 22, 2011 Why Personalize? 12 The Paradox of Choice by Barry Schwartz Sunday, July 17, 2011
  • 13. IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
  • 14. IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
  • 15. IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
  • 16. IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
  • 17. IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser Sunday, July 17, 2011
  • 18. IJCAI, Industry Day Barcelona :: July 22, 2011 Personalization is great ...but 13 The Filter Bubble by Eli Pariser IMPORTANT UNCOMFORTABLE CHALLENGING OTHER POINTS OF VIEWS Sunday, July 17, 2011
  • 19. IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 14 A Recommender selects the product that if acquired by the buyer maximizes value of both buyer and seller at a given point in time Sunday, July 17, 2011
  • 20. IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Sunday, July 17, 2011
  • 21. IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Recommender Components Sunday, July 17, 2011
  • 22. IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Knowledge Base Recommender Components Sunday, July 17, 2011
  • 23. IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Knowledge Processing Application Knowledge Base Recommender Components Sunday, July 17, 2011
  • 24. IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components Sunday, July 17, 2011
  • 25. IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components Sunday, July 17, 2011
  • 26. IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 25%25%25%25% Sunday, July 17, 2011
  • 27. IJCAI, Industry Day Barcelona :: July 22, 2011 What is a Recommender? 15 A Recommender processes information and transforms it into actionable knowledge User Interface It has a certain level of autonomy presenting recommendations to the end user Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 25%25%25%25% Sunday, July 17, 2011
  • 28. IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 16 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
  • 29. IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 Sunday, July 17, 2011
  • 30. IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 1. Do I need a recommender? Sunday, July 17, 2011
  • 31. IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 1. Do I need a recommender? 2. How do I get one? Sunday, July 17, 2011
  • 32. IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 1. Do I need a recommender? 2. How do I get one? 3. Is it performing well? Sunday, July 17, 2011
  • 33. IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 17 1. Do I need a recommender? 2. How do I get one? 3. Is it performing well? 4. Was it a good idea after all? Sunday, July 17, 2011
  • 34. IJCAI, Industry Day Barcelona :: July 22, 2011 Agenda 18 About Strands What is a Recommender? The Business Perspective Top Lessons Learned Sunday, July 17, 2011
  • 35. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 19 Sunday, July 17, 2011
  • 36. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 20 Sunday, July 17, 2011
  • 37. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 21 Sunday, July 17, 2011
  • 38. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 22 X X X X X 000s in both products and customers low medium high lowmediumhigh DiversityoftheCatalog Diversity of the Customers OK Sunday, July 17, 2011
  • 39. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
  • 40. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
  • 41. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
  • 42. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
  • 43. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
  • 44. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
  • 45. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
  • 46. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 1 Make sure it is needed. 23 lowmediumhigh ROI showing this impact is already challenging! random top 10 recommender sophisticated recommender even more sophisticated recommender Sunday, July 17, 2011
  • 47. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 24 Sunday, July 17, 2011
  • 48. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 25 Is the best recommendation for the customer the best recommendation for the business? Sunday, July 17, 2011
  • 49. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 26 Relevant vs Useful Correctness is often too obvious to be useful Riskier recommendations have less chances of being known customer perspective business perspective Short-term gain vs long-term return Sunday, July 17, 2011
  • 50. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % Sunday, July 17, 2011
  • 51. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % Sunday, July 17, 2011
  • 52. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % How much business logic goes into Recommender? Sunday, July 17, 2011
  • 53. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % How much business logic goes into Recommender? Whats the right level of autonomy a recommender must have? Sunday, July 17, 2011
  • 54. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 2 It must makestrategicsense. 27 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base ? %? % How much business logic goes into Recommender? Whats the right level of autonomy a recommender must have? How can the business control recommendations? Sunday, July 17, 2011
  • 55. IJCAI, Industry Day Barcelona :: July 22, 2011 The Business Perspective 28 1. Do I need a recommender? 2. How do I get one? 3. Is it performing well? 4. Was it a good idea after all? Sunday, July 17, 2011
  • 56. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 29 Sunday, July 17, 2011
  • 57. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Sunday, July 17, 2011
  • 58. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Small company Select a vendor Sunday, July 17, 2011
  • 59. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Small company Select a vendor Medium company Hire a copule of PhD students! Sunday, July 17, 2011
  • 60. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Small company Select a vendor Medium company Hire a copule of PhD students! Large company Partner with an experienced vendor Sunday, July 17, 2011
  • 61. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 30 Small company Select a vendor Medium company Hire a copule of PhD students! Large company Partner with an experienced vendor Tech company Organize a contest! Sunday, July 17, 2011
  • 62. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 31 2000 2002 2003 2004 2006 2007 2008 20092001 20102005 Sunday, July 17, 2011
  • 63. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 31 2000 2002 2003 2004 2006 2007 2008 20092001 20102005 000s licensing + 000s integration Price $99 / month Sunday, July 17, 2011
  • 64. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 3 Choose the right partner. 31 2000 2002 2003 2004 2006 2007 2008 20092001 20102005 Vendors hundreds vendors few vendors 000s licensing + 000s integration Price $99 / month Sunday, July 17, 2011
  • 65. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 4 Cold start? Be creative! 32 Sunday, July 17, 2011
  • 66. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 4 Cold start? Be creative! 33 With the advent of the Internet the start for a Recommender isnt so cold anymore Sunday, July 17, 2011
  • 67. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 5 Data and algorithms. 34 Sunday, July 17, 2011
  • 68. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 5 Data and Algorithms. 35 Which really makes the difference? Ingredients or Receipe? Sunday, July 17, 2011
  • 69. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 5 Data and Algorithms. 36 X X bad good badgood DataQuantityandQuality Algorithm Performance OK Sunday, July 17, 2011
  • 70. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 6 Finding correlated items is easy, deciding what, how, and when to present to the user is hard. 37 Sunday, July 17, 2011
  • 71. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 6 Finding correlated items is easy, deciding what, how, and when to present to the user is hard. 38 Math Art Sunday, July 17, 2011
  • 72. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 7 Dont wast time calculating nearest neighbors. 39 Sunday, July 17, 2011
  • 73. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 7 Dont waste time calculating nearest neighbours. 40 Let people tell you! Sunday, July 17, 2011
  • 74. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 8 Dont wait too long to get ready to scale. 41 Sunday, July 17, 2011
  • 75. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 8 Dont wait too long to get ready to scale. 42 When is the right moment? if you do too soon and recommendations dont take off... if you do too late and recommendations do take off... Sunday, July 17, 2011
  • 76. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 43 Sunday, July 17, 2011
  • 77. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 44 Sunday, July 17, 2011
  • 78. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 45 Sunday, July 17, 2011
  • 79. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 46 Sunday, July 17, 2011
  • 80. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 47 Implicit Ratings vs Explicit Ratings Implicit Semantics vs Explicit Semantic of Ratings Sunday, July 17, 2011
  • 81. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 48 The Ideal (explicit) Rating system... feedback Good Badso-so* *optional at it may help to confirm some implicit actions. actions I have it (i knew it, i saw it) Dont show it any more Show it to me later has it (knew it, saw it)my friend Sunday, July 17, 2011
  • 82. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 9 Choose the right feedback mechanism. 49 Sunday, July 17, 2011
  • 83. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 10 Measure everything. 50 Sunday, July 17, 2011
  • 84. IJCAI, Industry Day Barcelona :: July 22, 2011 LESSON 10 Measure everything. 51 4th ACM Conference on Recommender Systems Barcelona :: September 30, 2010 Sunday, July 17, 2011
  • 85. IJCAI, Industry Day Barcelona :: July 22, 2011 52 LESSON 10 Measure everything. Sunday, July 17, 2011
  • 86. IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components Sunday, July 17, 2011
  • 87. IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 50% Sunday, July 17, 2011
  • 88. IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 50%20% Sunday, July 17, 2011
  • 89. IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 50%20%5% Sunday, July 17, 2011
  • 90. IJCAI, Industry Day Barcelona :: July 22, 2011 So, what have we learned? 53 User Interface Business Control & Analytics Knowledge Processing Application Knowledge Base Recommender Components 50%20%5%25% Sunday, July 17, 2011
  • 91. IJCAI, Industry Day Barcelona :: July 22, 2011 Questions? 54 Thank you! Sunday, July 17, 2011