Link-based features
In this exercise, you will compute first order link-based features by multiplying the Churn
attribute of the network with the network's adjacency matrix.
Note, that since churn is a binary indicator, the attribute Churn
has 1 for churners and 0 for non-churners. Consequently, the attribute 1-Churn
has 1 for non-churners and 0 for churners.
This is helpful when computing the number of non-churn neighbors.
This exercise is part of the course
Predictive Analytics using Networked Data in R
Exercise instructions
- Compute the attribute
ChurnNeighbors
, i.e. the number of neighbors who churned, by multiplyingAdjacencyMatrix
with theChurn
attribute ofnetwork
. Applyas.vector()
to the result and add it to the network. - Similarly, compute
NonChurnNeighbors
, i.e. the number of non-churn neighbors. - Calculate the attribute
RelationalNeighbor
, the ratio of churners in the neighborhood, by dividingChurnNeighbors
with the sum ofChurnNeighbors
andNonChurnNeighbors
.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
# Compute the number of churn neighbors
V(network)$ChurnNeighbors <- as.vector(___ %*% V(network)$___)
# Compute the number of non-churn neighbors
V(network)$___ <- as.vector(___ %*% (1 - V(network)$___))
# Compute the relational neighbor probability
V(network)$RelationalNeighbor <- as.vector(V(network)$___ /
(V(network)$___ + V(network)$___))