Studying Diffusion of Viral Content at Dyadic Level

Transcript

1. Studying Diffusion of Viral Content at Dyadic Level ASONAM, SSNAD 2012 ASONAM, SSNAD 2012 Anita Zbieg, Błażej Żak, Jarosław Jankowski, Radosław Michalski, Sylwia Ciuberek Institute of Psychology, University of Wrocław, Wrocław, Poland, Faculty of Computer Science, West Pomeranian University of Technology, Szczecin, Poland Faculty of Computer Science and Mgmt., Wrocław University of Technology, Wrocław, Poland 1
2. Talk outline 1. Inspiration for the study 2. Research area 3. Theoretical framework 4. The study 4. The study 5. Results 6. Conclusions 2
3. Inspiration for the study 3
4. I. Research area 4
5. Related works • The more connections seeding individual had, the smaller information cascade has been observed, and viral content received from individuals with many ties had bigger chance to be ignored – Liu-Thompkins, Yuping, Seeding Viral Content: Lessons from the Diffusion of Online Videos. Journal of Advertising from the Diffusion of Online Videos. Journal of Advertising Research. • Seeding strategies are the most effective if aimed at well-connected individuals: hubs with many connections, or bridges that connect different groups – O. Hinz, B. Skiera, Ch. Barrot, J.U. Becker, Seeding Strategies for Viral Marketing: An Empirical Comparison, Journal of Marketing 6 . 75 (2011) : pp. 55-71 5
6. Related works • People with many connections seem to spread viral content less effectively (than those with few ties) and are more difficult to activate because they receive information from many because they receive information from many other sources – Liu-Thompkins, Yuping, Seeding Viral Content: Lessons from the Diffusion of Online Videos. Journal of Advertising Research. 6
7. II. Theoretical framework Theoretical framework 7
8. Study scope • Levels of network analysis –Individual actor level of analysis –Dyads level of analysis • two actors and their ties – Triad level of analysis • three actors and their ties – Subgroup level of analysis – Global level of analysis Source: Wasserman and Faust (1994) 8
9. Dyad as a basic unit of analysis 9 Communcation event
10. Basic behavior patterns investigated „Dead end” „Broadcasting” 10 „Diffusion fuel” „Brokerage”
11. Edge labelling 11
12. III. The study The study 12
13. Two time periods • T1 – 5 days before diffusion (dyad metrics) • T2 – 5 days of diffusion (behavior patterns) T1 T2 Collect metrics Observe behaviors Diffusion starts 13
14. Network & information cascade T1 T2 Color: Number of nodes 2362 Number of edges 25134 Average node degree 10.64 Nodes in the cascade 324 14
15. Any differences between observed behavior motifs (dyads)? • Mann-Whiney U test – Non parametric statistical test – Does not expect the distribution of – Does not expect the distribution of observed variables to follow the normal distribution (e.g. power law) – Availible in most statistical packages 15
16. IV. Results Results 16
17. Dyads where receiver made use of the virtual item VS Dyads where reciever did NOT use the virtual item No significant differences 17 Sender & receiver: activity, degree, betweenness, eigenvector, authority, reputation, age. Relationship: interactions, neighborhood overlap (common friends) .
18. Dyads that ended cascade VS propagated the message Significant differences Significant differences ↓ Influencer authority ↑ ↓ Relation interactions ↑ ↓ Relation neighborhood overlap ↑ ↓ Influenced activity ↑ ↓ Influenced degree ↑ ↓ Influenced betweenness ↑ ↓ Influenced eigenvector ↑ 18
19. Finding characteristics of particular behavior motif vs others (D2 ∪ D3 ∪ D4) vs others (D1 ∪ D3 ∪ D4) vs others (D1 ∪ D3 ∪ D4) vs others (D1 ∪ D2 ∪ D4) vs others (D1 ∪ D2 ∪ D3) 19
20. Sender (influencer) n Activity Reputation Degree Betweenn ess Eigenvector Authority Age D1 vs (D2 ∪ D3 ∪ D4) 89 - - - - - - - D2 vs 156 - - - - - - - vs (D1 ∪ D3 ∪ D4) - - - - - - - D3 vs (D1 ∪ D2 ∪ D4) 146 - - - - - - - D4 vs (D1 ∪ D2 ∪ D3) 22 - - ↑ * - ↑ * ↑ * - * significance level 0.05, ** significance level 0.01 20
21. Relationship (channel) n Interactions Neighborhood overlap D1 vs (D2 ∪ D3 ∪ D4) 89 - - D2 vs (D1 ∪ D3 ∪ D4) 156 - ↓ ** D3 146 D3 vs (D1 ∪ D2 ∪ D4) 146 ↑ * ↑ ** D4 vs (D1 ∪ D2 ∪ D3) 22 - - * significance level 0.05, ** significance level 0.01 21
22. Receiver (influenced) n Activity Reputation Degree Between ness Eigenvector Authority Age D1 vs (D2 ∪ D3 ∪ D4) 89 - - - - - - - D2 vs (D1 ∪ D3 ∪ D4) 156 ↓ ** ↓ ** ↓ ** ↓ ** ↓ ** - - D3 vs (D1 ∪ D2 ∪ D4) 146 ↑ ** ↑ ** ↑ ** ↑ ** ↑ ** ↑ ** ↓ * D4 vs (D1 ∪ D2 ∪ D3) 22 - ↑ * - - - - - * significance level 0.05, ** significance level 0.01 22
23. V. Conclusions Conclusions 23
24. If you want to start a protest (or spread a viral message)… Engage close friends (active relationship and many common friends) that are the most active and central in the network. 24
25. Conclusion A position that person occupies in a network determines in some way the role she/he plays in a diffusion process - If she/he receives a viral information - If she/he transfers a viral information to friends and other people 25
26. Conclusion How information is propagated in many ways depends on a network we are embedded in – information received from authority is more likely to be shared further. to be shared further. However the decision about using it (at least in this study) seems to be individual and independent from the network 26
27. Thank you! Thank you! 27