Method for predicting churners in a telecommunications network

1. A method for identifying potential churning telephony customers, said method comprising the steps of:

obtaining data pertaining to telephone calls between a plurality of customers;

forming a graph having a plurality of nodes representing said customers and a plurality of edges representing said telephone calls between said customers;

denoting a sub-set of said customers as previously churned customers; and

applying a spreading activation model to said graph to identify, based on said graph and said previously churned customers, said potential churning customers, wherein applying said spreading activation model comprises:

activating a subset of said nodes of said graph;

assigning a spreading factor and a transfer function specifying a fraction of energy transferred by a given one of said subset of said nodes to corresponding neighboring nodes and a distribution of said transferred energy among said neighboring nodes; and

ceasing said iterative steps when said subset of activated nodes does not substantially change between successive iterations and when changes in energy between iterations for a given node are less than a predetermined accuracy threshold.

2. The method of claim 1 , wherein said potential churning customers are identified as individuals.

3. The method of claim 1 , wherein said potential churning customers are identified as groups.

4. The method of claim 1 , wherein said step of applying said spreading activation model comprises the sub-step of applying weights to each of said plurality of edges based upon said nodes and said edges.

5. The method of claim 4 , wherein said graph further comprises subgraphs and wherein said weights are further based upon said subgraphs.

6. The method of claim 5 , wherein said weights are based upon said subgraphs by linearly scaling said weights of said edges by a factor D, where said factor D comprises density of a given one of said subgraphs, said density of said given one of said subgraphs in turn comprising a ratio of a number of edges for said given one of said subgraphs to a number of nodes for said given one of said subgraphs.

7. The method of claim 5 , wherein:

said step of applying said spreading activation model comprises performing a plurality of iterative steps; and

said spreading activation model is based upon a plurality of parameters, said parameters in turn being based upon said nodes, said edges, and said subgraphs, said parameters comprising:

initial energy of nodes associated with said previously churned customers;

a fraction of energy propagated by each affected node to its neighboring nodes;

a normalization function to determine a substantially exact amount of energy that each of said neighboring receives from a corresponding one of said affected nodes; and

an accuracy threshold for predicting a given one of said affected nodes as a node corresponding to a given one of said potential churning customers;

wherein said affected nodes comprise, in a first one of said iterative steps, direct neighbor nodes of said previously churned customers, and in subsequent ones of said iterative steps, direct neighbor nodes of nodes which were ones of said affected nodes in an immediately previous one of said iterative steps.

8. The method of claim 1 , further comprising the additional step of applying at least one additional churn prediction mechanism to said data, wherein said identification of said potential churning customers is based on said graph, said previously churned customers, and said at least one additional churn prediction mechanism.

9. The method of claim 8 , wherein said spreading activation model and said at least one additional churn prediction mechanism are integrated using at least one of a set intersection and a set union.

10. The method of claim 1 , further comprising the additional step of facilitating contact with at least a fraction of said potential churning customers to reduce a probability of said potential churning customers actually churning.