STOCHASTIC PROCESSES AND QUEUEING THEORY TERM PAPER ON

SOCIAL NETWORKING

ANSIL S SHAJIL
EE16M010
Social group is a dynamic entity with a structure that changes over time.It reaches its statistical
equilibrium that is transition between states continue to occur but a group is in a specific
state reaches a limiting constant value. The structural development of a group is modelled as a
finite state Markov chain.

Let X(t) be a binary matrix valued stochastic process with elements X_ij(t)
where
Xij(t)
= 1 if i chooses j at time t

0 otherwise

Consider a specific arc Xij(t) . If Xij(t)=1 then the individuals i and j are related at the time instant t
and Xij(t)=0 means they are not related at the time instant t. It will remain in that state for an
exponentially distributed amount of time. Its average waiting time is the reciprocal of the
parameter of this exponential function is a complex function of the entire network .During the
time interval t to t+h ,the binary value x evaluates the network and decides whether to change
or not.

The Basic Modelling Framework

The two assumptions

1. Markov
X(t) is a standard markov chain with finite state space S and probability transition function
Px,y(t,h).

Px,y(t,h)=P
[X(t+h)=y | X(t)=x]

2.Conditional Change Independence

Px,y(t,h)=

i,j (P[ Xi,j(t+h)=
yi,j|X(t)=x]
) +o(h)

In the small interval of time (t,t+h) relationship can only change in two ways
If there is no connection between individual i and j at time t . But there is a bond between
the two at time t+h.
i and j are connected at time t .But the bond does not exist at time t+h.

1
They depend on the current state x and time t.

Simple Stochastic Models

1.Reciprocity Model

0ij(x,t)=
0x
0+ ji

1ij(x,t)=
1x
1+ ji

0ij and 1ij do not depend on time (transition rates are time homogeneous and the
network stochastic process is stationary in time). The transition rates depend upon xji by
a linear equation with coefficients 0 and 0 or 1 and 1.

If there are n no of individuals , then there is a possibility of n(n-1) no of relations and

n(n-1) pairs of transition rates for the reciprocity model. But for all these , rate of
relationships between i and j only depends upon the relation between j and i .

We define dyad Dij(t) as for the pair of nodes (i,j). Each dyad is independent of all other
dyads .
Di,j=(
Xij(t),
Xji(t)
)

2
 M(t)=
P[ Di,j(t)=(1,1)]
(mutual)
A1(t)=
P[ Di,j(t)=(1,0)]
(asymmetric)
A0(t)= P[ Di,j (t)=(0,1)] (asymmetric)
N (t)= P[ Di,j (t)=(0,0)] (null)

Be the elements of (t), the vector of probabilities of the four state dyad process. If we
let t ,( ) is the vector of equilibrium probabilities which are

2.Popularity model

The popularity model has transition states

Individual is choice or non choice of individual j depends only on the number of

individuals who chose j, xj. Like the reciprocity model , the transition rates are time

3
homogenous and network process time homogenous. There are also n(n-1) pairs of
rates with popularity model but for a fixed j , the n(n-1) pairs for various i are identical.
That is each individual in the group has a unique pair of transition rates that are
constant for all unique pair of transition rates that are constant for all choices to be
made of him or her.There are only n distinct pairs of rates. The choices made of
individual j do not depend on the number of choices received by the other group
members. Indegree is a measure of the social status of the group members. The
popularity model is an attempt to postulate how the status of an individual within a group
influences the choice made of him or her.

The parameters 0 and 1 are measures of the overall rate of change for the group. The
popularity parameters 0 and 1 measure the importance of popularity of individual j to
the relationships of group members to j . since a large indegree should increase the
tendency of j to be chosen by i ,0 is positive and unrestricted by magnitude.
Now consider 1. A large indegree should decrease the chance of choices disappearing
. Hence 1 <0.

Let Xj(t) = (X1j(t), X2j(t),...........Xnj(t)) be the jth column process of the digraph process
X(t) . each of the g column processes has 2g-1 states consisting of all possible zero-one
vectors of length g with Xjj(t)=0. Xj(t) is independent and identically distributed . the
entire digraph process can be represented as g independent column processes
consisting of columns of X(t) each a continuous-time 2g-1 state Markov chain.
Notice the block-diagonal appearance of the generator . This block diagonal feature is
true for all values of g. The structure of the generator for general g:0+(k-1)0 above the

4
diagonal and 1+(k+1)1 below the diagonal where k is the number of ones in the
vectors contained in the diagonal block. The remaining off-diagonal elements in the
matrix are zero .

The popularity matrix is identical to a model for expansiveness with parameterization .

The expansiveness model reduces X(t) to be a set if g independent and identically

distributed row processes consisting of the rows of X(t). These row processes are
mathematically quivaet to the column processes with 0 and 1, the parameters
measuring the importance of the willingness to make many choices replacing 0 and 1
in all formulas.

By examining the indegree process derived from the column process , we can compute
the equilibrium probabilities of Xj(t).

Define

As the indegree process , the sum of the number of ones in the jth column process at
time t . It is a continuous birth-death process since by assumption an indegree can only
increase by one , decrease by one or remain same in a small interval of time. We know
that the equilibrium probabilities of the indegree process are

Ck
Conditional on k , the number of ones in a column vector , every one of the g-1
vectors with k ones and g-1-k zeros are equally likely .The probability mass function of

5
the column process is just a multiple of the probability mass function of the indegree
process; specifically

For k=0,1(n-1) where xj has k ones and (g-1-k) zeros as the equilibrium
probabilities of Xj. The column process can be analyzed by means of the simpler
indegree process Ij(t).

6
7