Project October Final Presentation

The final presentation describing October, a project for EECS395 at Case Western Reserve University.

"Read the news you want to read"

Transcript

1. October Tom Dooner, Brian Stack, Mika Little
2. Eternal September
3. “Read the news you want to read”
4. Frontend Ruby on Rails, PostgreSQL Backend Finagle (Scala), MongoDB API Apache Thrift
5. (demo)
6. Frontend
7. +
8. { 34 => 0.032491, 42 => 0.023807 }
9. Featured Article
10. Primary Article
11. Secondary Article
12. Users
13. (demo)
14. Thrift
15. MockThrift x
16. Backend
17. +
18. Tokens, Posts, Posters
19. Tokens, Posts, Posters
20. The Recommendation Algorithm:
21. 1. Get a user’s keyword list {“cwru”: 130, “cleveland”: 19, “pandas”: 13}
22. 2. Get a list of recent posts containing those tokens [post 1, post 2, post 3]
23. 3. Normalize keyword frequencies using TF-IDF { “cwru”: 0.8842, “cleveland”: 0.4519, ...}
24. 4. Calculate the dot product against the user’s vector. For each post, { “cwru”: 0.8842, “sustainability”: 0.4519, ...} x {“cwru”: 130, “cleveland”: 19, “pandas”: 13}
25. 5. Take the top posts and merge them in order by weight with recommendations for people you follow.
26. 6. Scale weights by time w’ = w * ln(age - 10 hours)-1
27. 7. Return list to frontend!
28. Backend Backend Backend ... {
30. https://october.case.edu Version 1.0 - April 15th, 2013
31. 43 users 108 recommendations
32. 0 25 50 75 100 < 1 sec 1 - 2 sec > 2 sec Recommendation Time
33. 69.7% of users click a headline
34. Thanks! • Code: https://github.com/ProjectOctober • Thanks to Grace Eder for the splash page stock photo