Copy-mutate processes for growth of bipartite networks: an application to cultural evolution Osame Kinouchi, Antônio Carlos Roque Adriano de Jesus Holanda (FFCLRP-USP) Rosa Wanda Diez Garcia (FMRP) Pedro Zambianchi (Faculdade Bandeirantes)
Why culinary? Physicists like to explain and model interesting statistical patterns (power laws) New application of complex networks ideas Database relatively easy to construct Analogy to biological evolution: growth network algorithm Similar to other humans, physicists like to eat
Why cookbooks? Recipes = cultural replicators (memes) Recipes in standard algorithmic form Cookbooks provide judicious (non-random) selection of recipes
Ingredients and Recipes Bipartite network
Cultural invariance
Temporal invariance
Complementary Cumulative Distribution
Recipe degree distribution
Copy-mutate model
Fitness selection Fitness fi interval [0,1] Substitute if fj > fi Recipe with K ingredients fi Fitness fi interval [0,1] Substitute if fj > fi fj Ingredient pool
Model results
Founder Effect
Founder Effect (II)
Fitness functions
Slow dynamics 1-F(t) = c t-g
Conclusions Bipartite network of ingredients and recipes is scale free with non-trivial exponent a 1.7 (out of range of generalized Yule process) Rank-frequency plot can be fitted by Darwinian copy-mutate process with ingredient fitness Model suggests presence of “founder effect” and slow (glassy) dynamics in cultural evolution
Acknowledgments: CNPq
Rank Entropy
Rank Entropy