The roundtable Segregation in conversation dynamics Massimo Mastrangeli

The roundtable Segregation in conversation dynamics Massimo Mastrangeli, Lucas Lacasa and Martin Schmidt Complex Systems Summer School 2009

Prelude: modeling conversation dynamics What drives social conversations? How do conversations evolve? �A realistic model needs to include: �Aggregation vs. protagonism balance �Individuality �Topology �Memory �Abstract from conversation content �Simple, but not too simple 07/03/2009 CSSS 09, Santa Fe Institute 2

Main theme: baseline model – boundary conds �Fellows (= FSA) sitting on a 1 D torus, each connected to their 2 first neighbours �Each fellow has a time-dependent happiness level �Each fellow addresses the others according to his own preference (= probability density function) �Fixed number of fellows �Uniform initial population (same happiness and pdf for all) involved in the same (unique) conversation �One speaker at the time �Segregation leads to multiple conversations 07/03/2009 CSSS 09, Santa Fe Institute 3

Main theme: baseline model – iterative dynamics �First speaker chosen randomly, other fellows silent �For each turn: �Speaker: 1) resets his happiness to the maximum, 2) chooses the new speaker among the fellows still in the conversation, 3) updates his own pdf accordingly �Silents decrease their happiness by 1 �Fellows who reach minimum happiness become latents, they look at their neighbours: If no latents -> remain latent 2. If part of other conversation -> joint the conversation 3. If latent -> start new conversation In 2) and 3) they become in and reset their happiness 1. 07/03/2009 CSSS 09, Santa Fe Institute 4

Main theme: first impressions �Propositions: �The model has a (degenerate) equilibrium state: �For N even: (N/2) 2 -subchats �For N odd, (N/2 – 1) 2 -subchats + (1) 3 -subchat �The rate of approach to the equilibrium state depends on: �Maximum happiness / N ratio �Probability density distributions �Memory levels �Topology �Proofs: �Ehm… the margins of this page are too narrow 07/03/2009 CSSS 09, Santa Fe Institute 5

Reprise: more investigations and insights �Analytical and computational investigations �Good old “pen and paper” �Fortran (fixed point analysis) �Netlogo (ABM) �Further: �Effects of: preferences, noise �Generalized topologies �Node degree > 2 �Cocktail party (agents moving and interacting in bounded space) 07/03/2009 CSSS 09, Santa Fe Institute 8

Coda: Wait and see… The Complex Social Segregation Sect would like to thank for their support: Tom Carter Daniel Rockmore … and all the wonderful crowd of CSSS 09 07/03/2009 CSSS 09, Santa Fe Institute 9