Conserved capacity of app usage - Data Science Lecture

Transcript

1. Strategies and limitations in app usage and human mobility Marco De Nadai (http://www.marcodena.it)
3. 3 Mobility, Social interactions are conserved
4. 4 Mobility, Social interactions are conserved
5. 5 What about apps?
6. 6 Photo: Liz Hafalia, The Chronicle + NBC News
7. 7 THERE ARE 2.1M APPS IN THE GOOGLE PLAY STORE App usage statistics by Statista.com & hubspot.net
8. 8 PEOPLE SPEND AN INCREASINGLY AMOUNT OF TIME ON THE PHONE […] REACHING 3 HOURS PER DAY THERE ARE 2.1M APPS IN THE GOOGLE PLAY STORE App usage statistics by Statista.com & hubspot.net
9. 9 PEOPLE SPEND AN INCREASINGLY AMOUNT OF TIME ON THE PHONE […] REACHING 3 HOURS PER DAY THERE ARE 2.1M APPS IN THE GOOGLE PLAY STORE App usage statistics by Statista.com & hubspot.net USER ENGAGMENT INCREASED BY 23%
10. 10 But I always use the same apps!
11. 11 ALWAYS USED But I always use the same apps!
12. 12 ALWAYS USED But I always use the same apps!
13. 13 ALWAYS USED But I always use the same apps!
14. 14 ALWAYS USED MEH... SOMETIMES But I always use the same apps!
15. 15 ALWAYS USED MEH... SOMETIMES But I always use the same apps!
16. 16 ALWAYS USED MEH... SOMETIMES WAS THIS APP EVEN HERE? But I always use the same apps!
17. 17 ALWAYS USED MEH... SOMETIMES WAS THIS APP EVEN HERE? • APP USAGE CHANGES • APPS CHANGE But I always use the same apps!
18. 18 ALWAYS USED MEH... SOMETIMES WAS THIS APP EVEN HERE? • APP USAGE CHANGES • APPS CHANGE But I always use the same apps!
19. 19 Monotony vs Growth of app usage
20. 20 •Screen time of 90K people •Over 8 months •69K different apps •12M of people’s hours spent on apps Data
21. 21 Data (pseudo-anonymized)
22. 22 Data (pseudo-anonymized) •Screen time •No network-based approaches •Unique dataset of app usage
23. 23 • ! ~ ( + " )#$#! " • = 1.19 ± 0.01 Characterizing App usage
24. 24 • ! ~ ( + " )#$#! " • = 1.19 ± 0.01 Characterizing App usage • ! ~ ()#$ • = 1.27 ± 0.01
25. 25 • ! ~ ( + " )#$#! " • = 1.19 ± 0.01 Characterizing App usage • ! ~ ()#$ • = 1.27 ± 0.01 • ~ % • γ = 0.41
26. 26 • ! ~ ( + " )#$#! " • = 1.19 ± 0.01 Characterizing App usage • ! ~ ()#$ • = 1.27 ± 0.01 • ~ % • γ = 0.41 • PEOPLE TIME IS FOCUSED ON FEW APPS • PEOPLE KEEP EXPLORING NEW APPS
27. 27 The familiar apps over time: the App space THE APP SPACE (t)
28. 28 • Apps used at least twice and for at least 10 min per week • 20 weeks long sliding windows (1 week) • How does the App space evolve over time? The familiar apps over time: the App space THE APP SPACE (t) THE APP SPACE (t-1) THE APP SPACE (t+1) 1 WEEK 1 WEEK
29. App space over time (adopted and dropped apps) THE APP SPACE (t)
30. App space over time (adopted and dropped apps) ! THE APP SPACE (t)
31. App space over time (adopted and dropped apps) ! ! = (!) = 4 THE APP SPACE (t)
32. App space over time (adopted and dropped apps) " ! ! = (!) = 4 " = (") = 5 THE APP SPACE (t)
33. App space over time (adopted and dropped apps) # " ! ! = (!) = 4 " = (") = 5 # = (#) = 4 THE APP SPACE (t)
34. App space over time (adopted and dropped apps) !$" = !$" = "$# = "$# = # " ! ! = (!) = 4 " = (") = 5 # = (#) = 4 THE APP SPACE (t)
35. 35 • Gain: 3 = 3 − 3 () • 97.5% of people exhibit a conserved capacity ( 4# 5$# ≤ 1) Adoped and dropped apps
36. What is the conserved capacity? On average: •Everytime a new app is adopted, an older app is discarded THE APP SPACE (t)
37. What is the conserved capacity? On average: •Everytime a new app is adopted, an older app is discarded THE APP SPACE (t)
38. What is the conserved capacity? On average: •Everytime a new app is adopted, an older app is discarded THE APP SPACE (t)
39. What is the conserved capacity? On average: •Everytime a new app is adopted, an older app is discarded THE APP SPACE (t)
40. 40 •Local shuffle: app usage order of a user •If the random permutated user has the same capacity => Time •Two-sample Kolmogorov–Smirnov test •Set p-value threshold (0.05 or 0.001?) - Different (p-value < threshold) - Don’t know (p-value > threshold) Is capacity a time-constraints consequence? ! " # $ USER A # $ ! " https://en.wikipedia.org/wiki/Kolmogorov%E2%80%93Smirnov_test
41. 41 •Local shuffle: app usage order of a user No time consequence (KS: 0.55 p-value < 0.001). •Global: app usage among users (KS: 0.98 p-value < 0.001). It’s an individual characteristics. Is capacity a time-constraints consequence? ! " # $ USER A # $ ! " ! " # $ USER A ! " # $ ! " # $ USER B ! " # $
42. 42 The app strategy • EXPLORERS (% ≫ ) • KEEPERS (% ≪ ) • Explorers adopt 1 new app every 28 weeks • Keepers always keep using the same apps % = % / %
43. 43 The app strategy • EXPLORERS (% ≫ ) • KEEPERS (% ≪ ) • Explorers adopt 1 new app every 28 weeks • Keepers always keep using the same apps % = % / %
44. 44 The app strategy • EXPLORERS (% ≫ ) • KEEPERS (% ≪ ) • Explorers adopt 1 new app every 28 weeks • Keepers always keep using the same apps % = % / %
45. 45 What about age?
46. 46 What about age?
47. 47 What about age? 16 What about age?
48. 48 People are focused on few apps, but keep exploring. Individuals exhibit a conserved app capacity •Capacity is an individual behaviour •Capacity varies with age •We can define app explorers and keepers •AND MORE! So?
49. 49 Then?
50. 50 Then?
51. 51 Then?
52. 52 Apps, Mobility, Social interactions are conserved DUNBAR’S NUMBER OF APPS?
53. 53 Apps that stay the most ~10% of apps are always kept ~17.5% are continuosly changed
54. 54 Strategy vs age
55. 55 EXERCISEEEE www.menti.com – Code: 2145921
56. 56 EXERCISEEEE! shorturl.at/nwSZ8
65. 65 What did I learn?
66. 66 Individuals exhibit a conserved app capacity •To explore dataaa •Understand complex queries •A bit of spark What did I learn?
67. 67 Individuals exhibit a conserved app capacity •To explore dataaa •Understand complex queries •A bit of spark What did I learn?
68. 68 Individuals exhibit a conserved app capacity •To explore dataaa •Understand complex queries •A bit of spark What did I learn?
69. Thank you! @denadai2