simulating the social processes of science leiden| 9 april 2014
DESCRIPTION
www.ingenio.upv.es. Simulating the Social Processes of Science Leiden| 9 April 2014. DIFFUSION OF SCIENTIFIC KNOWLEDGE: A PERCOLATION MODEL Elena M. Tur With P Zeppini and K Frenken. INGENIO [CSIC-UPV] Ciudad Politécnica de la Innovación | Edif 8E 4º Camino de Vera s/n - PowerPoint PPT PresentationTRANSCRIPT
Simulating the Social Processes of Science Leiden| 9 April 2014
INGENIO [CSIC-UPV] Ciudad Politécnica de la Innovación | Edif 8E 4º Camino de Vera s/n 46022 Valencia
tel +34 963 877 048 fax +34 963 877 991
DIFFUSION OF SCIENTIFIC KNOWLEDGE:
A PERCOLATION MODEL
Elena M. Tur
With P Zeppini and K Frenken
www.ingenio.upv.es
2
• How do social pressure and assortativity affect the process of diffusion?
• How do their effect interact with the social network structure?
• How do local effects dictate global diffusion through the overall network structure?
• Main contribution: introducing social pressure in a model of diffusion with different network structures
Focus
2
3
• Spread of rumors (Zanette, 2002)• Online spreading of behavior (Centola, 2010)• Opinion dynamics (Shao et al., 2009)• Paradigm shift in science (Brock and Durlauf, 1999)
• How to approach the problem (Zeppini and Frenken, 2013)• Heterogeneity of agents• Externalities• Social network• Percolation (Solomon et al., 2000)
Diffusion of ideas
4
• : value of the idea• : minimum quality requirement of agent
Agent adopts the idea at time if:• has not adopted before • is informed: a neighbor has adopted at time • is willing to adopt:
Percolation in a social network
5
Percolation in a social network
𝑞𝑖∼𝑈 [0,1 ]→𝑞𝑖≤𝑣0?
6
Percolation in a social network
7
• Small world (Watts and Strogatz, 1998)Starting with a regular lattice, with nodes and 𝑛 𝑘edges per node, rewire each edge at random with probability 𝑝
• High clustering coefficient, low average path length
Social network
8
• Upper bound to diffusion: 45º line (well-mixed population)
• Diffusion increases with • Phase changes: from a
non-diffusion to a diffusion regime
• Percolation thresholds decrease with
Percolation without social pressure
9
SOCIAL PRESSURE
10
• The unwilling to adopt can be persuaded • Network externalities: shared language, shared
experiences, shared beliefs…• Increasing amount of evidence• Conformity (Asch,1958)
• Weariness
Why social pressure?
11
• : # neighbors of agent that have adopted at time • : social pressure intensity
• is decreasing in the number of adopting neighbors • is decreasing in the social pressure intensity• if • if
Modelling social Pressure
12
• agents• original neighbors in the small world algorithm• seeds or original adopters• time periods• runs of Monte Carlo simulations
• iid• Social networks: lattice, small world, Poisson network
Parameters for the simulations
12
13
Percolation with social pressure
14
Clustering with social pressure
16
ASSORTATIVITY People tend to be friends with similar people
17
Modelling assortativity
Most unwilling to adopt
Most willing to adopt
18
• No critical transition from non-diffusion to diffusion
• Diffusion size scales linearly with diffusion size
• No effect of the network structure (well-mixed population)
Percolation with assortativity
19
Percolation with social pressure and assortativity
20
Does assortativity help diffusion?
21
Does assortativity help diffusion? (2)
22
CONCLUSIONS
23
• Social pressure changes the behavior of the percolation process: higher diffusion size, lower percolation thresholds
• The effect of social pressure is different for different network structures: critical transition in small worlds and lattices
• With social pressure, clustering becomes beneficial for diffusion: reduced differences between networks
• Assortativity removes the effect of the network structure: all behave as a well-mixed population
• Assortativity can help or hamper diffusion depending on the setting
Conclusions
24
• Other social network structures: scale-free networks (Barábasi and Albert, 1999)
• Evolving endogenous social network• Re-think the social pressure implementation: always
possitive efffect?• Different minimum quality requirement distributions,
Future work and limitations
25
THANK YOUINGENIO [CSIC-UPV] Ciudad Politécnica de la Innovación | Edif 8E 4º Camino de Vera s/n 46022 Valencia
tel +34 963 877 048 fax +34 963 877 991