Netflix recommendation systems

The Netflix recommendation systems (technical and marketing)
for the university.

1. What is recommendation systems?

AMK
[email protected]

2. Recommender systems definition

3. We use
recommendation
systems
everywhere

4. Recommender
systems
purpose

5. The origin of the story : 1990

6. Profits:
1.Helps users to make the decision
2.Increases users awareness in the field of interest

7. Data collection

8. Target shops story

9. Patterns

10. Type of recommendation systems:
1.Collaborative filtering
2.Content based filtering
3.Hybrid systems

11. Collaborating filtering :

12. Content based filtering :

13. Summary :

14. how does recommendation system
Influence on marketing and sales?
Navid AZ
[email protected]
Case:

15. What is?

16. Reed Hastings and Marc Randolph co-found Netflix to offer online movie rentals;
1997

17. The company launched the first DVD rental and sales site , netflix.com

1998 – 2000

18. 4.2 million members

20. 2007

Netflix introduced streaming

21. Netflix launched its first slate of original
programming
2013
-the first internet TV network nominated for the Emmy Awards

26. 2016

Netflix is available worldwide: 190 countries;

27. 2017

Netflix 100 million members globally
- Netflix wins its first Oscar for Best Documentary Short Subject

30. 2018

The most nominated service at 2018 Emmy Awards with 112
nominations (HBO received 108); both Netflix and HBO won 23
awards each.


In year!

31. Paid memberships – over 148 million in 190 countries;

Revenue in Q1 2019 – over $4.5 billion (Netflix);

Stock-market value – nearly $165 billion, more than Disney ($130
billion);

Netflix has raised a total of $3.1 billion in funding over 10 rounds;
NOW

33. Marketing Funnel

34. Business Model
subscription-based
●Basic: 9$ per month, single screen at a time, SD
●Standard: 13$ per month, two screen at a time, HD
●Premium: 16$ per month, four screen at a time, 4K

35. 8 key factors behind Netflix’s success story

36. 1. Disrupting the industry – Disrupting itself

37. 2. Flexibility

38. 3. Investing in original content

41. 4. No ads

42. 5. Netflix pioneered binge-watching

●54% market size for netflix
●10% screen time in the US

43. 6- Clever Campaigns

47. 7. Excellent user experience

-Custom-created preview videos that
automatically play when you scroll over
a title card;

-A download-and-go feature that
allows users to watch shows offline;

-The ability to share your Netflix
account;

-30-day free subscription.

48. 8. Personalization via the Netflix recommendation engine

49. “Almost everything we do is a recommendation.”
-Netflix engineering director Xavier Amatriain

51. ●Investors
●Content Owners
●Internet Service Providers
●Filmmaker Guides and Individuals
●Cinemas and Theaters
●Prizes and Film Festivals
●Influencers
●IP Holders
●Pornhub!
●TV Network Companies
●Google and Amazon
●Alliances with Smart TV Companies
●Alliances with the Gaming Industry

Netflix different partners

52. How does
recommendation system work?
(Matrix Factorization)
Mina Tafreshi
Case: Netflix technical review

53. Recommendation systems are one of the most fascinating
applications of machine learning!
But the question is:
-HOW do these work?

54. Answer:
Matrix factorization

55. Netflix Universe
Anna
Bob
Copper Dana
M2
M1 M3
M4 M5

56. Netflix ratings:


M1M2

M3

M4

M5

Anna1 3 2 5 4
Bob 2 1 1 1 5
Copp
er
3 2 3 1 5
Dana2 4 1 5 2

57. Human behave:


M1M2

M3

M4

M5

Anna1 3 2 5 4
Bob 2 1 1 1 5
Copp
er
3 2 3 1 5
Dana2 4 1 5 2

58. Human behave:


M1M2

M3

M4

M5

Anna3 1 1 3 1
Bob 1 2 4 1 3
Copp
er
3 1 1 3 1
Dana4 3 5 4 4
Anna = Copper

M1 = M4

Bob + Copper = Dana

M5 = Avg (M2, M3)

59. Human behave:


M1M2

M3

M4

M5

Anna3 3 3 3 3
Bob 3 3 3 3 3
Copp
er
3 3 3 3 3
Dana3 3 3 3 3
= = =

60. Guessing rate:


M1M2

M3

M4

M5

Anna3 3 3 3 3
Bob 3 3 3 3 3
Copp
er
3 3 3 ? 3
Dana3 3 3 3 3
= = =
Copper = 3 star for M4

61. How do figure out all these dependencies?

Answer :
matrix factorization

62. Factorization

6 * 4 = 24






M1M2

M3

M4

M5

Anna3 1 1 3 1
Bob 1 2 4 1 3
Copp
er
3 1 1 3 1
Dana4 3 5 4 4
this * that =

63. features
Comedy
Dreama
Action
Scary

Dot product
Has a sad dog
Good canadian
Meryl Streep

Anna
M1
TrueFalse


31

ActionComedy

64. Dot product
Dana
M1
TrueTrue


31

ActionComedy So this is how we do:

3 + 1 = 4

65. M1 3 1
M2 1 2
M3 1 4
M4 3 1
ComedyAction
AnnaTrueFalse
BobFalseTrue
CopperTrueFalse
DanaTrueTrue
Categories
M5 1 3
ActionComedy
Anna movie 1

3 + 1 = 3

66. M1M2

M3

M4

M5

Anna3 1 1 3 1
Bob 1 2 4 1 3
Copp
er
3 1 1 3 1
Dana4 3 5 4 4

67. 6 * 4 = 24

68. M1 M2

M3

M4

M5

Comedy 3 1 1 3 1
Action 1 2 4 1 3
ComedyAction
AnnaTrue False
BobFalse True
CopperTrue False
DanaTrue True


M1 M2

M3

M4

M5

Anna 3 1 1 3 1
Bob 1 2 4 1 3
Copper 3 1 1 3 1
Dana 4 3 5 4 4
matrix
factorization

69. storage

70. Matrix factorization
1000
movies
2000 user
2000 * 1000
2M entries
1000
movies
100 Features
2000 user
100 Features
100 * 1000
100k entries
2000 * 100
200k entries

71. How to find right factorization?
Training models

74. www.github.com/minatafreshi
[email protected]

75. Thanks for listening to me! :D