Carnegie Mellon School of Computer Science Modeling Website Popularity Competition in the Attention-Activity Marketplace Bruno Ribeiro Christos Faloutsos Carnegie Mellon University WSDM 2015 February 5, 2015

Bruno Ribeiro Carnegie Mellon School of Computer Science 2 Motivation  Bought for $580 million in 2005  Sold for $35 million in 2011 *source: TechCrunch MySpace’s demise Sold Hi5 Friendster Summer 2008

Bruno Ribeiro Carnegie Mellon School of Computer Science 3 Predicting Network Popularity w/ Competition Goal: Why: ? ? ? Summer 2008 Fraction of Active Users (t) -$545 million dollars

Bruno Ribeiro Carnegie Mellon School of Computer Science new adopters/semester 4 MySpace.com until late 2008 No Structural Telltale Signs of MySpace’s Demise Matches known theoretical behavior Mansfield’61, Rogers’03, Bass’69 Total Adopters And no abrupt topological changes in 2008 MySpace graph until recently: 35 million users ★

Bruno Ribeiro Carnegie Mellon School of Computer Science Economics: ◦ (Mansfield ’63) ◦ (Katz&Shapiro’85) ◦ (Farrell&Saloner ’86) ◦ (Choi ’94) ◦ (Arthur ’94) Marketing: ◦ (Bass ’69) ◦ (Fisher&Pry ’71) Fraction of Active Users (t) Summer 2008 Background: Vast Adoption Literature 5 Computer Science:  (Kempe et al ’03)  (Zhao et al., IMC’12)  (Leskovec et al., SIGKDD’08)  (Ugander et al., PNAS’12)  (Aral&Walker, Science’12) Sociology: o (Ryan&Gross’49) o (Everett ’62, ’03) o (Rogers ’03) o (Centola ’12) Adopt ≠ Active We know how to model adoption But how to model attention?

Bruno Ribeiro Carnegie Mellon School of Computer Science  H. A. Simon on information overload:  Information consumes attention (time) ◦ Information-rich world  Attention-poor world  Systems that “talk” more than “think” exacerbate information overload 6 Attention & Information Overload

Bruno Ribeiro Carnegie Mellon School of Computer Science  Facebook attention & activity 7 How Facebook Talks (a lot)

Bruno Ribeiro Carnegie Mellon School of Computer Science My Attention My Content Friends’ Attention Friends’ Content 8 Positive & Negative Attention Loops Positive Growth Negative Growth Website Survives Website Dies

Bruno Ribeiro Carnegie Mellon School of Computer Science 9 Proposed Competition Model

Bruno Ribeiro Carnegie Mellon School of Computer Science 10 Proposed State Space states ∅y∅y UyUy AyAy IyIy ∅x∅x UxUx AxAx IxIx Both x & y Only y for nowOnly x for now Neither for nowNot interested in x/y User Subscribes to: Website Killers Never!UnawareActiveInactive

Bruno Ribeiro Carnegie Mellon School of Computer Science 11 Model Transitions User Competition

Bruno Ribeiro Carnegie Mellon School of Computer Science 12 Isolated Network Transitions states ∅y∅y UyUy AyAy IyIy SUM ∅x∅x UxUx S (U, ∅ ) (t) AxAx S (A, ∅ ) (t)S (A,*) (t) IxIx S (I, ∅ ) (t) Media/Marketing gets new subscribers Word of mouth gets subscribers Network x activity attracts user back Other activity takes user away from x Details

Bruno Ribeiro Carnegie Mellon School of Computer Science Model Predictions with Constant Distraction Factor 13  Predicting Online Social Network Popularity ◦ How websites do on their own ◦ Evolution with competition Outline

Bruno Ribeiro Carnegie Mellon School of Computer Science Signatures of Self-Sustainability  Long-term User Activity [Ribeiro’14] 14 Relative attractiveness of activity Fraction of Active Users (t) t t ≅1≅1

Bruno Ribeiro Carnegie Mellon School of Computer Science Signatures of Popularity Growth  Model prediction: Signatures of activity growth [Ribeiro’14] 15 t t Fraction of Active Users (t) Word of mouth Media & Marketing

Bruno Ribeiro Carnegie Mellon School of Computer Science Predictions with Changing Distractions 16 t  Predicting Online Social Network Popularity ◦ How websites do on their own ◦ Evolution with competition Outline

Bruno Ribeiro Carnegie Mellon School of Computer Science  Two observable states: S (A,*) (t) and S (*,A) (t)  Remaining states latent & treated as parameters  Data from Alexa.com  Fitted with Levenberg–Marquardt algorithm using squared error 17 Model Fitting S (A,*) (t) S (I,*) (t)? S (U,*) (t)? S (A,I) (t)? S (A,A) (t)? S (A,U) (t)? S (A, ∅ ) (t)? S (I, ∅ ) (t)? S (U, ∅ ) (t)? friendster.com Fraction of Active Users (t) Details

Bruno Ribeiro Carnegie Mellon School of Computer Science MySpace Friendster Multiply Hi5 18 Facebook Introduces Wall Distraction of Concurrent Users Increases as Facebook Introduces Wall Distraction of Concurrent Users Increases as Facebook Introduces Wall

Bruno Ribeiro Carnegie Mellon School of Computer Science 19 Results (Facebook x MySpace) t value

Bruno Ribeiro Carnegie Mellon School of Computer Science 20 Results (Facebook x Multiply) t value Multiply

Bruno Ribeiro Carnegie Mellon School of Computer Science 21 Results (Facebook x Hi5) t value Hi5

Bruno Ribeiro Carnegie Mellon School of Computer Science 22 Results (Facebook x Hi5) Model Captures Coexistence and Then Death of Facebook Competitors Too Many Concurrent Users = Fragility Model Captures Coexistence and Then Death of Facebook Competitors Too Many Concurrent Users = Fragility

Bruno Ribeiro Carnegie Mellon School of Computer Science  Attention feedback helps predict / understand network survival  No marketing can save from negative attention loop  Insights from model ◦ E.g.: Metrics of node centrality should take attention feedback into account 23 Conclusions

Bruno Ribeiro Carnegie Mellon School of Computer Science Thank you! Questions? 24

Bruno Ribeiro Carnegie Mellon School of Computer Science 25 Attention Competition Transitions states ∅y∅y UyUy AyAy IyIy SUM: ∅x∅x UxUx S (U,A) (t) AxAx S (A,U) (t)S (A,A) (t)S (A,I) (t ) S (A,*) (t) IxIx S (I,A) (t) SUM:S (*,A) (t) Both x & y Only y for now Only x for now Neither for now Media/Marketing new subscribers Word of mouth subscribers Not interested in x/y Only parameter affected by attention competition “Geek factor” User Subscribes to: Details