Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Do Diffusion Protocols Govern Cascade Growth?

Do Diffusion Protocols Govern Cascade Growth?

Large cascades can develop in online social networks as people share information with one another. Though simple reshare cascades have been studied extensively, the full range of cascading behaviors on social media is much more diverse. Here we study how diffusion protocols, or the social exchanges that enable information transmission, affect cascade growth, analogous to the way communication protocols define how information is transmitted from one point to another. Studying 98 of the largest information cascades on Facebook, we find a wide range of diffusion protocols – from cascading reshares of images, which use a simple protocol of tapping a single button for propagation, to the ALS Ice Bucket Challenge, whose diffusion protocol involved individuals creating and posting a video, and then nominating specific others to do the same. We find recurring classes of diffusion protocols, and identify two key counterbalancing factors in the construction of these protocols, with implications for a cascade’s growth: the effort required to participate in the cascade, and the social cost of staying on the sidelines. Protocols requiring greater individual effort slow down a cascade’s propagation, while those imposing a greater social cost of not participating increase the cascade’s adoption likelihood. The predictability of transmission also varies with protocol. But regardless of mechanism, the cascades in our analysis all have a similar reproduction number (≈1.8), meaning that lower rates of exposure can be offset with higher per-exposure rates of adoption. Last, we show how a cascade’s structure can not only differentiate these protocols, but also be modeled through branching processes. Together, these findings provide a framework for understanding how a wide variety of information cascades can achieve substantial adoption across a network.

Presented at ICWSM 2018.

Justin Cheng

June 26, 2018
Tweet

More Decks by Justin Cheng

Other Decks in Research

Transcript

  1. Justin Cheng, Jon Kleinberg, Jure Leskovec, David Liben-Nowell,
    Bogdan State, Karthik Subbian, and Lada Adamic

    View Slide

  2. i.e., how do memes go viral?
    how
    (not why)

    View Slide

  3. View Slide

  4. View Slide

  5. Prior work
    Cha, et al. (2009, 2010); Romero, et al. (2011); Goel, et al. (2012); Weng, et al. (2013); Cheng, at al. (2014, 2016)
    cascade growth prediction
    cascade recurrence
    contagion/diffusion models
    network structure
    network evolution
    social influence

    View Slide

  6. A focus on reshare cascades
    Prior work
    Hubs crucial to growth
    Simple replication
    Goldenberg, et al. (2009); Cheng, at al. (2014)
    cascade growth prediction
    cascade recurrence
    contagion/diffusion models
    network structure
    network evolution
    social influence

    View Slide

  7. Types of cascades
    (This figure for illustrative purposes only.)
    Reshare Cascades
    Other Cascades

    View Slide

  8. What other types of cascades exist?

    View Slide

  9. View Slide

  10. Reshare Post

    View Slide

  11. Reshare Post Ice Bucket Challenge Post

    View Slide

  12. View Slide

  13. 17M videos
    200K participants
    2M supporters

    View Slide

  14. Do different social mechanisms
    influence cascade growth?
    Research Question
    diffusion protocols
    Explain variations in cascade growth

    View Slide

  15. What is a diffusion protocol?

    View Slide

  16. Inspiration: Communications Protocols
    Client Server
    1. SYN
    2. SYN-ACK
    3. ACK

    View Slide

  17. Inspiration: Communications Protocols
    Client Server
    1. SYN
    2. SYN-ACK
    3. ACK

    View Slide

  18. Inspiration: Communications Protocols
    Client Server
    1. SYN
    2. SYN-ACK
    3. ACK

    View Slide

  19. Inspiration: Communications Protocols
    Client Server
    1. SYN
    2. SYN-ACK
    3. ACK

    View Slide

  20. View Slide

  21. Transient Copy Protocol

    View Slide

  22. Transient Copy Protocol

    View Slide

  23. View Slide

  24. Transient Copy Protocol

    View Slide

  25. Nomination Protocol
    Transient Copy Protocol

    View Slide

  26. Nomination Protocol
    Transient Copy Protocol

    View Slide

  27. Nomination Protocol
    Transient Copy Protocol

    View Slide

  28. View Slide

  29. Transient Copy
    Diffusion Protocols
    Nomination

    View Slide

  30. Transient Copy Persistent Copy Nomination Volunteer
    Diffusion Protocols

    View Slide

  31. Transient Copy Persistent Copy Nomination Volunteer
    Also see: State and Adamic (2015)
    Diffusion Protocols

    View Slide

  32. Transient Copy Persistent Copy Nomination Volunteer
    Diffusion Protocols

    View Slide

  33. Transient Copy Persistent Copy Nomination Volunteer
    Diffusion Protocols

    View Slide

  34. Transient Copy Persistent Copy Nomination Volunteer
    Each protocol can generate large cascades!
    ~1M adoptions
    per cascade
    ~800K adoptions
    per cascade
    ~6M adoptions
    per cascade
    ~1M adoptions
    per cascade

    View Slide

  35. Diffusion protocols
    govern cascade growth
    Our Hypothesis
    Individual effort Social cost

    View Slide

  36. Participation requires individual effort
    Marwell, Oliver, and Prahl (1988)

    View Slide

  37. Participation can have social costs
    Milgram (1969); Granovetter (1978); Banerjee (1992)

    View Slide

  38. Participation can have social costs
    Cialdini (1993); Goffman (1967)

    View Slide

  39. How does individual effort and
    social cost explain differences?

    View Slide

  40. Cascades on Facebook
    98 large cascades 200M users
    4 protocols
    (All data was de-identified and analyzed in aggregate.)

    View Slide

  41. How do cascades of different
    diffusion protocols vary on…
    1 Propagation Speed
    2 Reliance on Hub Nodes
    3 Tie Strength
    4 Simple or Complex Diffusion

    View Slide

  42. How do cascades of different
    diffusion protocols vary on…
    1 Propagation Speed
    2 Reliance on Hub Nodes
    3 Tie Strength
    4 Simple or Complex Diffusion

    View Slide

  43. How quickly do these
    cascades propagate?

    View Slide

  44. Persistent Copy Nomination
    % Users
    How quickly do cascades propagate?
    26
    21 211 216
    0%
    12%
    24%
    0%
    5%
    10%
    15%
    Transient Copy Volunteer
    Adoption Delay (s)
    221 26
    21 211 216 221
    0%
    8%
    16%
    0%
    8%
    16%

    View Slide

  45. Persistent Copy Nomination
    % Users
    Copy cascades spread quickly…
    26
    21 211 216
    0%
    12%
    24%
    0%
    5%
    10%
    15%
    Transient Copy Volunteer
    Adoption Delay (s)
    221 26
    21 211 216 221
    0%
    8%
    16%
    0%
    8%
    16%

    View Slide

  46. Persistent Copy Nomination
    % Users
    …while non-copy cascades spread slowly
    26
    21 211 216
    0%
    12%
    24%
    0%
    5%
    10%
    15%
    Transient Copy Volunteer
    Adoption Delay (s)
    221 26
    21 211 216 221
    0%
    8%
    16%
    0%
    8%
    16%
    ~ 1 day

    View Slide

  47. Transient Copy Nomination
    Increasing effort reduces propagation speed
    Less effort?
    More effort?

    View Slide

  48. Transient Copy Volunteer
    Increasing effort reduces propagation speed
    2
    1
    3
    4
    1

    View Slide

  49. How do cascades of different
    diffusion protocols vary on…
    1 Propagation Speed
    2 Reliance on Hub Nodes
    3 Tie Strength
    4 Simple or Complex Diffusion

    View Slide

  50. Are hubs always crucial in the
    development of large cascades?
    Goldenberg, et al. (2009); Hinz, et al. (2011); Cheng, et al. (2014)

    View Slide

  51. Persistent Copy Nomination
    How significant are hub nodes?
    Transient Copy Volunteer
    % Adoptions Originating from the Top 1% of Users
    10
    0 20 30 40 10
    0 20 30 40

    View Slide

  52. Persistent Copy Nomination
    Low-effort cascades benefit more from hubs
    Transient Copy Volunteer
    % Adoptions Originating from the Top 1% of Users
    10
    0 20 30 40 10
    0 20 30 40
    13.7 7.6
    5.8
    37.3

    View Slide

  53. Hubs Non-Hubs
    Katona, et al. 2011
    Hubs are less influential per individual

    View Slide

  54. How do cascades of different
    diffusion protocols vary on…
    1 Propagation Speed
    2 Reliance on Hub Nodes
    3 Tie Strength
    4 Simple or Complex Diffusion

    View Slide

  55. Does tie strength vary
    across cascades?

    View Slide

  56. Persistent Copy Nomination
    How does tie strength vary?
    Transient Copy Volunteer
    # Mutual Friends
    20
    0 40 60 80 20
    0 40 60 80

    View Slide

  57. Persistent Copy Nomination
    Transient copy cascades use weaker ties…
    Transient Copy Volunteer
    # Mutual Friends
    20
    0 40 60 80 20
    0 40 60 80
    28.8

    View Slide

  58. Persistent Copy Nomination
    …while other cascades use stronger ties
    Transient Copy Volunteer
    # Mutual Friends
    20
    0 40 60 80 20
    0 40 60 80
    28.8 44.6
    58.1
    70.7

    View Slide

  59. Transient Copy Nomination
    Social costs invoke strong ties
    Lower
    social cost
    Greater
    social cost

    View Slide

  60. Transient Copy Persistent Copy
    Social costs invoke strong ties
    Demonstrates
    group
    belonging

    View Slide

  61. How do cascades of different
    diffusion protocols vary on…
    1 Propagation Speed
    2 Reliance on Hub Nodes
    3 Tie Strength
    4 Simple or Complex Diffusion

    View Slide

  62. Is the diffusion process
    simple or complex?
    Centola and Macy (2007)

    View Slide

  63. Persistent Copy Nomination
    P(Adopt)
    Is diffusion simple or complex?
    3
    1 5 7
    Transient Copy Volunteer
    # Prior Exposures
    9 3
    1 5 7 9
    0
    0.003
    0.006
    0.009
    0
    0.002
    0.004
    0.006
    0
    0.06
    0.12
    0.18
    0
    0.003
    0.006
    0.009

    View Slide

  64. Persistent Copy Nomination
    P(Adopt)
    Social costs can increase adoption rates
    3
    1 5 7
    Transient Copy Volunteer
    # Prior Exposures
    9 3
    1 5 7 9
    0
    0.002
    0.004
    0.006
    0
    0.003
    0.006
    0.009
    0
    0.003
    0.006
    0.009
    0
    0.06
    0.12
    0.18

    View Slide

  65. Persistent Copy Nomination
    P(Adopt)
    Complexity depends on cascade type
    3
    1 5 7
    Transient Copy Volunteer
    # Prior Exposures
    9 3
    1 5 7 9
    0
    0.002
    0.004
    0.006
    Complex
    0
    0.003
    0.006
    0.009
    Simple
    0
    0.003
    0.006
    0.009
    0
    0.06
    0.12
    0.18

    View Slide

  66. Transient Copy Persistent Copy
    Social proof leads to complex diffusion

    View Slide

  67. Individual effort and social cost
    influence cascade growth

    View Slide

  68. Transient Copy Persistent Copy Nomination Volunteer

    View Slide

  69. High
    Indiv. Effort
    Low
    Indiv. Effort
    Low Social Cost
    High Social Cost
    Persistent Copy Nomination
    Transient Copy Volunteer

    View Slide

  70. High
    Indiv. Effort
    Low
    Indiv. Effort
    Low Social Cost
    High Social Cost
    Hub-Centric
    Strong Ties
    Complex Diffusion
    Fast
    Persistent Copy
    Not Hub-Centric
    Strong Ties
    Simple Diffusion
    Slow
    Nomination
    Hub-Centric
    Weak Ties
    Simple Diffusion
    Fast
    Transient Copy
    Not Hub-Centric
    Strong Ties
    Complex Diffusion
    Slow
    Volunteer

    View Slide

  71. How do cascades of different
    diffusion protocols vary on…
    1 Propagation Speed
    2 Reliance on Hub Nodes
    3 Tie Strength
    4 Simple or Complex Diffusion

    View Slide

  72. How do cascades of different
    diffusion protocols vary on…
    1 Propagation Speed
    2 Reliance on Hub Nodes
    3 Tie Strength
    4 Simple or Complex Diffusion
    5 Adoption Rate

    View Slide

  73. How do adoption rates
    vary across cascades?

    View Slide

  74. How do adoption rates vary?
    (Considering only non-leaf nodes)
    % Users Adopted
    0%
    15%
    30%
    45%
    60%
    # Users Exposed
    0 75 150 225 300

    View Slide

  75. How do adoption rates vary?
    (Considering only non-leaf nodes)
    % Users Adopted
    0%
    15%
    30%
    45%
    60%
    # Users Exposed
    0 75 150 225 300
    Ice Bucket Challenge cascade

    View Slide

  76. How do adoption rates vary?
    (Considering only non-leaf nodes)
    % Users Adopted
    0%
    15%
    30%
    45%
    60%
    # Users Exposed
    0 75 150 225 300
    Transient copy cascade
    Ice Bucket Challenge cascade

    View Slide

  77. Adoption rates drop as exposures increase
    (Considering only non-leaf nodes)
    % Users Adopted
    0%
    15%
    30%
    45%
    60%
    # Users Exposed
    0 75 150 225 300
    Transient Copy
    Persistent Copy
    Nomination
    Volunteer

    View Slide

  78. The reproduction number is constant!
    (Considering only non-leaf nodes)
    % Users Adopted
    0%
    15%
    30%
    45%
    60%
    # Users Exposed
    0 75 150 225 300
    y =
    1.81
    x

    View Slide

  79. Parallels with r/K selection theory
    Pianka (1970)
    K-selected Species
    r-selected Species
    Stayin’ alive!

    View Slide

  80. Parallels with r/K selection theory
    “K-selected” Protocol
    “r-selected” Protocol

    View Slide

  81. Parallels with r/K selection theory
    “K-selected” Protocol
    “r-selected” Protocol

    View Slide

  82. Large cascades can result from
    diverse diffusion protocols

    View Slide