Social-circles network feedback model

From InterSciWiki
Jump to: navigation, search

("Feedback networks")

This is a copy of the Wikipedia:Social-circles network model written by D.R. White. (Author(s) Doug White)

Generative model

The generative model of feedback networks [1], [2] studied by White, Kejžar, Tsallis, Farmer, and White, or social-circles network model, defines a class of random graphs generated by simple processes that are common to edge formation and feedback loops in social circles. The implications of this type of model are carried further in Thurner, Kyriakopoulos and Tsallis, 2007, Unified Model for Network Dynamics Exhibiting Nonextensive Statistics Phys. Rev. E 76, 036111 (2007) (8 pages). This class of models is distinct from the Wikipedia:small-world network and the Wikipedia:scale-free network models in Wikipedia:network analysis but subsumes many of their characteristics and also captures many of the characteristics of real-world Wikipedia:social networks.


  • A feedback network is one where active nodes or agents send tokens or communications through their network to help locate potential partners for new ties.
  • When a previously unconnected partner is located given constraints of distance a new tie is formed, creating a feedback loop.
  • Failing to locate a partner within the network, the active node engaged in partner selection recruits a new partner from outside the existing network and forms a tie.
  • Like many real-world networks, feedback networks modeled by this process evolve large networks with cohesive ties (social circles). In this way, small worlds with cohesive subgroups and relatively short distances may be generated, Wikipedia:scale-free networks may be generated, or a variety of other network topologies may be generated.
  • The generative feedback network model simulates these processes and their network outcomes as new nodes and edges are added to an initial network.
  • Only three parameters are used to govern network formation in this probabilistic generative model. These govern levels of node activity, distance decay in how soon network traversal fails to find a partner, and the extent to which local hubs are used in network traversal.
  • The model provides an explanation for how hubs are formed in networks, how well-traveled edges are formed, how cycles and cohesion are formed, and how other features of real-world networks may result from different combinations of three basic network parameters.
  • Empirical network data can be fitted to the parameters of the general model.


A social-circles network models by a graph-generating process the formation of cohesion and dispersion in connected networks as a function of activity and feedback. The process begins with a single node. The next events in the evolution of the network are that: (1) a node is chosen randomly (proportional to the number of current links of each node raised to power alpha) to emit a feedback token, (2) the token attempts to travel by a non-circular path through the current network by choosing its next neighbor randomly, proportional to the number of edges of the neighbor minus the node already traversed, raised to power gamma; (3) the token travels through the network until it reaches a random distance d, proportional to 1 over d raised to a power beta, and failing that, the source node that originated the token gains an edge to a node that is added to the network (dispersion of edges), but if the token succeeds in reaching a target at distance d, a new 'feedback' edge is added between the source and the target.

The three parameters of the model are alpha, beta, gamma. The notion of 'feedback' in this model is very general and may involve intentionality on the part of source nodes as agents looking for a partner or simply nodes in a network emitting signals randomly that contain information that may enable the source and target to lock into a feedback relation, represented by the formation of a new edge. Whether searches or signals have the capacity to locate a target in an existing network is affected by the current size and topology of the network, the Wikipedia:distance decay (beta) in traversal, and the intelligence (gamma) exhibited in the method of search. Failure to locate targets results in a generic substitution (new node and an edge connecting to it), like looking in the phone book for a doctor rather that asking for a referral.

The generative model

At each step three stochastic processes are run. A node i is selected, a distance d (an imaginary number of steps to get to a searched node j) is chosen, and the search path of distance d is generated.

A node i is selected with a probability proportional to its degree
p_\alpha(i)=\frac{[deg(i)]^\alpha}{\sum_{m=1}^N [deg(m)]^\alpha} (activity)
A distance d is chosen with the probability
p_\beta(d)=\frac{d^{-\beta}}{\sum_{m=1}^\infty m^{-\beta}} (distance decay)
where d represents the number of steps necessary to create a new cycle. If the number of steps cannot be concluded, a new node with a new edge is addded.
A search path of length d is generated. At a given moment a node r is a current node on the path. The next node is selected among the neighbors of r that were not yet visited. The probability of selecting a certain neighbor (say l) is proportional to their 'unused degree'. The 'unused degree', u(l), is considered a degree of the node from which all the already visited neighbors are subtracted. The probability is given by
p_\gamma(l)=\frac {[1+u(l)^{\gamma}]} {\sum_{m=1}^N[1+u(m)^{\gamma}]} (hub use)


Social-circles network simulation with varying model parameters results in a variety of networks whose topology in terms of network density, clustering, cohesion, and Wikipedia:degree distribution varies in ways that resemble the variety of real-world networks. Study of the degree distributions and edge-traversal frequencies shows fit to the statistical distributions of Wikipedia:Tsallis entropy, familiar to nonextensive physics. This provides a baseline test of the model to real-world networks, one that contrasts with that of the Wikipedia:scale-free network, for which one baseline test is whether the degree distribution follows a Wikipedia:power law.

The Wikipedia:Tsallis entropy distribution asymptotes to a power law for large degree but bends toward the Wikipedia:exponential distribution in the lower part of the degree distribution.

Degree Distributions

Parameters \alpha, \beta, \gamma in the social circles model are shown to generate networks with degree distributions that fit a general family of curves that combine power-law with exponential tendencies. The probability density function for degree k in this family, with parameters \delta, \kappa, q , and p_0, is given by:

p(k)=p_0k^\delta e_q^{-k/\kappa}

where the q-exponential function e^{-x}_q with parameter q is defined as

e^{-x}_q = (1+(1-q)x)^{1/(1-q)}   	(e_1^x = e^x)

The q-exponential function arises naturally as the solution of the equation dk/dt=k^q, which is k=e^{-x}_q = [1+(1-q) a t]^{1/(1-q)}. For the case where q=1, k=e^{at}. The function is related to the stationary solution of a nonlinear Fokker-Planck equation known as the porous medium equation. It describes correlated diffusion known as Wikipedia:Darcy's law, the scientific basis of fluid permeability used in the earth sciences. This is a constitutive equation that describes the flow of a fluid through a porous medium as formulated by Henry Darcy to describe results of experiments on the flow of water through beds of sand (see references [2,3,4]). In the density function, p_0 coincides with p(0) if and only if \delta=0; \kappa is a characteristic degree. In a generalization of the water-flow analogy, the density function for degree would reflect a q-exponential component for which q=1 is omnidirectional flow, and as q increases there is initially a steep low-dimensional gradient that changes to flat and high dimensional as q grows to infinity. The other component corresponding to p_0k^\delta would reflect a characteristic magnitude of flow also affected by q. A loose analogy is to traders whose activities etch the routes of their commercial networks, amplified by gradient intensities and the heterogeneity of terrain.


The parameters \delta,  \kappa,  q, and p_0 of the fitted models, which may also be applied to the degree distributions of empirical networks, were found to closely approximate 1-to-1 functions of the generating parameters, as given by the approximations

\delta = -1.5 \alpha
q = 1+.64\alpha + 6\alpha\gamma(\beta-1)^2 (a reasonable estimate if 1 \le  q \le 2).
\kappa = 235\gamma(1-\alpha)(\beta-1)
p_0 = 2 + \alpha/4  - \beta  -1.3\gamma

These allow approximate solutions for the generating parameters that fit observed empirical degree of networks:

\alpha = - 2\delta/3 (activity bias)
\beta = 3/2 + p_o/2 + q/9 (distance decay)
\gamma = 5|(q/k)-.5|^{1/2} + 5p0+ 11q/k (hub use)

The last of these approximations (for \gamma) is the least accurate (more precise estimates may yet be found). The authors of the paper have provided a more exact optimal matching between the four parameters fitted from empirical degree distributions of actual networks (assuming they follow the q-exponential) in spreadsheet form.[3] This makes the social circles model a prime candidate for accounting for the degree distributions of empirical networks in terms of a generative network model with the three parameters \alpha, \beta, \gamma: activity bias of agents, distance decay, and navigability of the network. Wikipedia:Maximum likelihood estimates (MLE) of the four \delta,  \kappa,  q, and p_0 parameters for empirical networks and Wikipedia:goodness-of-fit tests can establish the statistical relationship between model and data. Shalizi (2007) provides an MLE procedure for the Pareto II or q-exponential parameter fitting.[4]

Replication and extensions

Kejžar in 2007 replicated these results by simulating feedback networks up to N=5000 nodes, extended the findings to:

  1. Degree distributions for directed edges (indegree, outdetree). These were similar to those of the undirected model, especially for \gamma=0 but differed toward higher q when \gamma \rightarrow 1
  2. Densification. Growth of edges relative to nodes is initially linear, but soon changees to approximate a slightly superlinear power-law (for \alpha=0, \beta=1.6,  \gamma=0.5, e\approx 0.08+n^{1.08}, for example, where e are edge numbers and n numbers of node). Even with 1 billion nodes, however, this is scarcely more than an average degree of 5. Up to 6000 nodes, the average degree is less than two. For large degrees, however, the degree distribution approaches a power law with an exponent 1/(q-1) -\delta. Densification is a natural consequence.
  3. Average shortest paths. If l is the average shortest path and n the number of nodes, l shrinks with n if \gamma=1 but rises if \gamma=0 and a weighted sum of \alpha, \gamma determines the crossover (low \alpha, high \gamma has the most rapid shrinkage). In every case there tends to be a "shrinking limit" as n \to \infty
  4. Variance of shortest paths. Most of the shortest paths between nodes approach the "shrinking limit."

In short, the generative feedback model has realistic and nicely continuous characteristics, and these simulations show no evidence that the density function for degree k is not the correct function.


  1. ( arXiv:cond-mat/0508028 31 July 2005) "Generative Model for Feedback Networks" in Physical Review E, 016119 (2006, Douglas R. White, Nataša Kejžar, Constantino Tsallis, J. Doyne Farmer, Scott D. White. Final paper in pdf as SFI working paper. Reviewed 2005 in Europhysicsnews 36(6):218-220 by Stefan Thurner. There is a modification of the 2006 article that was published earlier by Thurner and Tsallis in 2005 (Thurner, Tsallis, and White were all at SFI together in 2005) as a physics model but the social science analogy would be that instead of acquiring a new partner, the last step in the White et al model would be that the "parter" would be join the searcher as one entity (e.g., a couple: man and wife). That article is S. Thurner and C. Tsallis. 2005. Nonextensive aspects of self-organized scale-free gas-like particles. Europhysics Letters 72(2): 197-203. Their abstract begins: "We explore the possibility to interpret as a "gas" the dynamical self-organized scale-free network recently introduced by Kim et al. (2005). (5 Mar 2004), See: Kim, Beom Jun, Ala Trusina, Peter Minnhagen and Kim Sneppen, 2005. Self organized scale-free networks from merging and regeneration. Eur. Phy. J. B 43, 2005. All three references {51, 52, 53) are given on p345 of Constantino Tsallis. 2009. [ Introduction to Nonextensive Statistical Mechanics: Approaching a Complex World. Springer. These articles come from a common thread of discussion in 2004-2005 stimulated by Tsallis.
  2. "Non-extensive statistical mechanics and generalized Fokker-Planck equation." 1995. A. R. Plastino and A. Plastino. Physica A 222, 347-354
  3. “Nonlinear diffusion equation, Tsallis formalism and exact solutions.’ 2005 P. C. Assis, Jr, L. R. da Silva, E. K. Lenzi, L. C. Malacarne, and R. S. Mendes Journal of Mathematical Physics 46, 123303 pdf.
  4. "Nonextensive statistical mechanics: A brief introduction", 2004. C. Tsallis and E. Brigatti. Continuum Mechanics and Thermodynamics 16(3):223-235. pdf
  5. Nataša Kejžar. 2007. Evolution of Networks. Doctoral thesis. Department of Mathematics. University of Ljubljana.


See also

Nigel Gilbert 2009 Social circles simulation

Abstract: Usually, the studies of distributions of city populations have been reduced to power laws. In such analyses, a common practice is to consider cities with more than one hundred thousand inhabitants. Here, we argue that the distribution of cities for all ranges of populations can be well described by using a q-exponential distribution. This function, which reproduces the Zipf-Mandelbrot law, is related to the generalized nonextensive statistical mechanics and satisfies an anomalous decay equation.
... In a comparative study, the q- exponential and Weibull distributions are employed to investigate frequency ...
  • Patriota, Alexandre G. 2012. A q-exponential regression model.
ABSTRACT: A critique of a non-extensive maximum entropy (NME) formalism is undertaken in conjunction with its application into the analysis of queues with heavy tails that are often observed in performance evaluation studies of heterogeneous networks exhibiting traffic burstiness, self-similarity and/or long range dependence (LRD). The credibility of the NME formalism, as a method of inductive inference, for the study of non-extensive systems with long-range interactions is explored in terms of four consistency axioms of extensive systems with short-range interactions. Focusing on a a general physical system and, as a special case, a single server queue with finite capacity, it is shown that the NME state probability is characterised by a generalisation of the Zipf-Mandelbrot (Z-M) type distribution depicting heavy tails and asymptotic power law behaviour. Typical numerical experiments are employed to illustrate the adverse combined impact of traffic burstiness and self-similarity on the behaviour of the queue. A reference to open issues relating to the NME formalism and open queueing networks is included.
KeywordsPerformance evaluation–extensive maximum entropy (EME) formalism–non-extensive maximum entropy (NME) formalism–generalised exponential (GE)–traffic burstiness–self-similarity–short-range dependence (SRD)–long-range dependence (LRD)–queueing systems–Zipf-Mandelbrot (Z-M) distribution

Pareto-Zipf-Mandelbrot distributions or Truncated Pareto-Zipf (slides)


  1. Cited by Wei, Wang, Qiuping, Nivanen, Lauret et al (2006-01-12) How to fit the degree distribution of the air network?
  2. Cited by de Meneses, Marcela, da Cunha, Sharon, and Soares et al (2006-01-12) Preferential attachment scale-free growth model with random fitness
  3. White et al. Retrofitting the Generative Feedback Network (Social Circles) Models
  4. Shalizi, Cosma. 2007. Maximum Likelihood Estimation for q-Exponential (Tsallis) Distributions.

fr:Analyse des réseaux sociaux