Your SlideShare is downloading. ×
How to build a Recommender System
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×

Introducing the official SlideShare app

Stunning, full-screen experience for iPhone and Android

Text the download link to your phone

Standard text messaging rates apply

How to build a Recommender System

3,686
views

Published on

This presentation show the method to build a Recommender System with Collaborative FIltering method. …

This presentation show the method to build a Recommender System with Collaborative FIltering method.

Published in: Technology

5 Comments
36 Likes
Statistics
Notes
  • @leminhhai that's cool :D
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • cũng không có biết anh Tuấn thuyết trình , nếu không ghé xem rùi
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • Về recommendation có thể tham khảo rất nhiều tài liệu hay từ http://www.netflixprize.com/ (Năm 2009 Netflix có tổ chức cuộc thi về recommendation, nếu nhóm nào tăng kết quả reommendation lên 10% thì được $1 triệu). Về tối ưu performance thì quả thực là vấn đề đau đầu, nếu có tiền thì dùng AMZ hay Google clould (+ sử dụng thêm Predition api). Có một bài viết khá hay về tính correlations 316 triệu phim của Bo Yang trong vòng 2 phút thay vì 2.5 giờ http://dmnewbie.blogspot.com/2009/06/calculating-316-million-movie.html . Một số recommendation service có thểm tham khảo là: http://www.recomaticapp.com/ http://monetate.com/ - expensive
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • kết thúc thiếu cái quan trọng là high performance thì làm như nào :D
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • Ôi, đợi cái Content-based filtering mà kết thúc cụt vậy
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
No Downloads
Views
Total Views
3,686
On Slideshare
0
From Embeds
0
Number of Embeds
5
Actions
Shares
0
Downloads
239
Comments
5
Likes
36
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Recommender System How to build a
  • 2. Võ Duy Tuấn Technical Director @ dienmay.com  PHP 5 Zend Certified Engineer  Mobile App Developer  Web Developer & Designer  Interest: o PHP o Large System & Data Mining o Web Performance Optimization o Mobile Development
  • 3. Introduction Collaborative Filtering Question & Answer AGENDA
  • 4. 1. Introduction
  • 5. APPLICATIONS • Personalized recommendation • Social recommendation • Item recommendation • Combination of 3 approaches above
  • 6. AMAZON.COM | BOOKS
  • 7. PLAY.GOOGLE.COM | APPS
  • 8. SKILLSHARE.COM | CLASSES
  • 9. PROCESS DIAGRAM Preprocessing Data Analysis Adjustment INPUT OUTPUT
  • 10. TYPE OF RECOMMENDER SYSTEM • Collaborative filtering • Content-based filtering • Hybrid
  • 11. 2. Collaborative Filtering
  • 12. USER & ITEM
  • 13. ORDER DATA
  • 14. ORDER DATA (cont.)
  • 15. ORDER DATA (cont.)
  • 16. VECTOR & DIMENSION
  • 17. VECTOR & DIMENSION
  • 18. VECTORS
  • 19. VECTORS
  • 20. SIMILARITY CALCULATION
  • 21. USER SIMILARITY MATRIX
  • 22. SIMILARITY CALCULATION
  • 23. SIMILARITY CALCULATION
  • 24. SIMILARITY CALCULATION EXAMPLE
  • 25. K-NEAREST-NEIGHBOR
  • 26. K-NEAREST-NEIGHBOR
  • 27. NEIGHBORS’ ORDER
  • 28. REMOVE BOUGHT ITEMS
  • 29. CALCULATING FINAL SCORE
  • 30. OTHER SIMILARITY MEASURES More at: http://favi.com.vn/wp-content/uploads/2012/05/pg049_Similarity_Measures_for_Text_Document_Clustering.pdf
  • 31. Problem ?!
  • 32. COLLABORATIVE FILTERING PROBLEM • Fail with cold start problem o New User o New Item • Performance o Large Data set o Pre-calculate
  • 33. PERFORMANCE EXAMPLE • We have 1,000,000 users (customers) • We sell 10,000 items - Total of similarity calculating = 1,000,000 x 1,000,000 = 1,000,000,000,000 - Each similarity calculate need 0.006s (on my MacBook Pro 2.2GHz Core i7, 8G Ram) => We need 1,000,000,000,000 x 0.006 = 6,000,000,000(s) ≈ 70,000 days ≈ 191 years - If store each similarity in 8 bytes, we need = 8,000,000,000,000 bytes ≈ 8,000 GB (on Memory or File)
  • 34. ITEM-TO-ITEM COLLABORATIVE FILTERING (AMAZON.COM ) Download Paper: http://www.cs.umd.edu/~samir/498/Amazon-Recommendations.pdf
  • 35. ADJUSTMENTS • Hybrid Recommender System • Sale forecast system • Context of User • Type of Item, Action • External (3rd-party) information.
  • 36. BOOKS Programming Collective Intelligence Toby Segaran Recommender Systems Handbook Many Authors Big Data For Dummies Marcia Kaufman, Fern Halper
  • 37. OPEN SOURCES
  • 38. Thank you! CONTACT ME: tuanmaster2002@yahoo.com 0938 916 902 http://bloghoctap.com/