What Happened on Tumblr? A Look into Tumblr as a Platform for Content Propagation

Transcript

1. What Happened on Tumblr?
   A Look into Tumblr as a Platform for
   Content Propagation
   Nora Alrajebah
   

   View Slide

2. Things I’ll cover today:
   1. Building Blocks
   2. Settings
   3. Findings
   4. Remarks
   

   View Slide

3. # 5JQTV *KUVQT[ QH 150U
   

   View Slide

4. The core affordances
   of OSNs
   

   View Slide

5. 1. Connect to others
   Follow
   Friend
   

   View Slide

6. 2. Create content
   

   View Slide

7. 5QQPCHVGT
   

   View Slide

8. 3. Express admiration
   favourite, like a.k.a. heart
   

   View Slide

9. 4. Spread content
   re += {tweet | blog | path} && share
   

   View Slide

10. View Slide

11. %CUECFGU
    

    View Slide

12. A cascade is considered as an artefact
    to study information diffusion on OSN
    It is the manifestation of the
    information diffusion process
    

    View Slide

13. Why analysing cascades is important?
    1. A proxy to unravel the way information is spread on the
    Web
    2. Explain the popularity-gaining phenomenon on the Web
    3. Estimate influence and homophily between users
    4. Estimate the value of the content
    5.They are better indicators of users’ interest and trust
    networks
    6.Explain social network evolution
    

    View Slide

14. Things I’ll cover today:
    1. Building Blocks
    2. Settings
    3. Findings
    4. Remarks
    

    View Slide

15. Tumblr
    • Diffusion on Tumblr is powered by
    the Reblogging functionality
    • The ability to reblog allow content
    to spread
    • The traces of the spread create
    cascades that can be observed
    • The reblogging events appears as
    a list of notes attached to posts
    and their reblogged copies
    

    View Slide

16. Content
    Unified, explicit and ordered
    

    View Slide

17. The two facets of a cascade
    Structural Temporal
    Who influenced whom to
    spread the content?
    How many shares are there at
    any point in time (Day/Hour)
    0!
    6!
    12!
    18!
    24!
    30!
    0! 2! 4! 6! 8! 10!
    Number of shares!
    Days after publishing!
    

    View Slide

18. The Web
    
    Social Interaction Data
    Sharing Data
    

    View Slide

19. The Web
    
    Social Interaction Data
    Sharing Data Other Interactions
    

    View Slide

20. The Web
    https://www.flickr.com/photos/frauhoelle/8464661409
    https://commons.wikimedia.org/wiki/File:Tumblr_icon.png
    
    Social Interaction Data
    Sharing Data Other Interactions
    5VTWEVWTCN 6GORQTCN 2NCVHQTO
    Explicit Cascades
    

    View Slide

21. Where do research tasks fit?
    Netw
    ork
    Science Data
    Science
    Web Science
    Cascades
    Construction
    Structural
    Analysis
    Data
    collection
    &
    preprocessing
    Temporal
    & Platform
    Analysis
    

    View Slide

22. Tumblr’s Year in Review
    “Tumblr’s Year in Review is a
    showcase of the best stuff on
    the Internet from 2014. Follow
    along for a daily dose of
    creativity, humor, humanity,
    fandom, and sharing.
    And GIFs. Lots of GIFs!”
    http://2014inreblogs.tumblr.com/2014
    

    View Slide

23. Stats
    Number of posts 1292
    No. or reblogs (rebloggers) 73,048,903
    No. of reblogees 3,541,110
    No. of likes 48,822,318
    No. of comments ??
    

    View Slide

24. Cascades networks construction
    • Ideally, cascade networks will have a neat tree topology ..
    • However ..
    • That is not the case, at least not all the time
    • This is due to cases where:
    • Reblogs are deleted
    • Users deactivate their accounts
    • Users reblog more than once
    Isolated
    components
    Repeated
    appearances of
    users
    

    View Slide

25. Things I’ll cover today:
    1. Building Blocks
    2. Settings
    3. Findings
    4. Remarks
    

    View Slide

26. Platform Analysis
    

    View Slide

27. High reblogging rate = Cascades are large!
    0 1 2 3 4 5 6
    x ⇥105
    0.0
    0.2
    0.4
    0.6
    0.8
    1.0
    P(Number of reblogs >= x)
    78%
    18%
    Yet another long tail on the Web!
    

    View Slide

28. But why?
    Short answer: reblogging is a cult
    Long answer: Tumblr employs numerous content
    exposure mechanisms
    

    View Slide

29. How ‘big’ are large cascades?
    

    View Slide

30. The Silent majority
    1.55% of reblogs are with comments, 0.32% with @
    There are 315 replies only
    0.16 comment for 10 reblogs
    

    View Slide

31. Reblogs > Likes
    A shamelessly committed community
    Their actions are stronger than their words
    7.9 like for 10 reblog
    

    View Slide

32. 0 2 4 6
    Cumulative sum ⇥107
    Reblogs
    Likes
    Comments
    

    View Slide

33. Structural Analysis
    

    View Slide

34. View Slide

35. 6YQRWTRQUGUŎ
    1- Influence estimators
    2- Structural estimators
    

    View Slide

36. Branching Factors
    i.e., the out-degree for each user
    

    View Slide

37. Q: How many users a user influences?
    0
    20
    40
    60
    80
    100
    %
    BF = 0 BF = 1 BF > 1
    ~68%
    ~12%
    ~20%
    

    View Slide

38. Subcascade sizes
    The size of the subcascade per user
    

    View Slide

39. Q: What is the overall impact of users?
    0
    20
    40
    60
    80
    100
    %
    Subcascade = 1 Subcascade > 1
    13%
    87%
    

    View Slide

40. Q: Small branching factor but high impact?
    - Compute:
    - Ratio = Branching factor / sub cascade size
    - if ratio > 1:
    - The user generates a subcascade that goes beyond
    its immediate effect, i.e. their branching factor
    - if ratio = 1:
    - The user generates subcascades that equal one
    

    View Slide

41. Q: What is the impact of the content’s author?
    0 20 40 60 80 100
    x %
    0.0
    0.1
    0.2
    0.3
    0.4
    0.5
    0.6
    P(Reblogs from author >= x%)
    Shallow
    Deep
    

    View Slide

42. Depth
    

    View Slide

43. Q: How far are the reblogging users from the
    post’s author?
    100 101 102
    Depth
    100
    101
    102
    103
    104
    105
    106
    No of Nodes
    Mean
    

    View Slide

44. Temporal Analysis
    

    View Slide

45. Posts Age
    • The eldest post was active for 617 days
    • The youngest was active for 28 days
    • Surprisingly, the old post has a very small cascade
    size of 131 reblogs, but still managed to survive for
    617 days! Accumulating popularity slowly but
    steady!
    

    View Slide

46. How long it takes a post to
    be reblogged?
    1 24
    Hours after publishing
    87% 97%
    

    View Slide

47. Growth patterns
    

    View Slide

48. how posts accumulate popularity?
    0 50 100 150 200 250 300 350
    Days after publishing
    0
    20000
    40000
    60000
    80000
    100000
    Reblogs per day
    school: 416282
    chill: 480379
    business: 371600
    17th
    

    View Slide

49. Q: Is there a cascade growth
    pattern on Tumblr?
    

    View Slide

50. 0.0 0.2 0.4 0.6 0.8 1.0
    Days after publishing
    0.0
    0.2
    0.4
    0.6
    0.8
    1.0
    Cascade size (cumulative)
    0.0 0.2 0.4 0.6 0.8 1.0
    Days after publishing
    0.0
    0.2
    0.4
    0.6
    0.8
    1.0
    Cascade size (cumulative)
    0.0 0.2 0.4 0.6 0.8 1.0
    Days after publishing
    0.0
    0.2
    0.4
    0.6
    0.8
    1.0
    Cascade size (cumulative)
    0.0 0.2 0.4 0.6 0.8 1.0
    Days after publishing
    0.0
    0.2
    0.4
    0.6
    0.8
    1.0
    Cascade size (cumulative)
    Q1 Q2
    Q3 Q4
    

    View Slide

51. Things I’ll cover today:
    1. Building Blocks
    2. Settings
    3. Findings
    4. Remarks
    

    View Slide

52. Wrapping up ..
    • Tumblr ‘year in review’ blog features some really ‘large’
    cascades!
    • Cascades matter!
    • Users’ influence might be underestimated if only the
    branching factor was taken into account
    • Cascades on Tumblr have non-trivial sizes and depths
    • Cascades grow in size in so many ways ..
    • Large cascades exist!
    

    View Slide

53. 6JCPMU
    

    View Slide