An ego network, or egonet, contains a focal node and its immediate neighbors. The focal node is the ego; surrounding nodes are alters. In graph feature engineering, this one-hop structure supplies local node features before broader network measures are considered. section 10.1
Use an ego network to derive progressively richer structural signals:
- Count the ego’s neighbors and inspect connections among them. section 10.1
- Count triangles involving the ego; classify them as fraudulent, legitimate, or semifraudulent when node labels are available. section 10.1.2
- Compute egonet density as observed internal edges divided by the maximum possible edges. section 10.1.3
- Aggregate local values recursively with ReFeX to capture increasingly distant structural context. section 10.3.1
Node A in a fraud-detection graph
Node A connects to B, G, H, I, O, and T, producing a seven-node egonet. It contains seven internal edges, nine edges leaving the egonet, density about 0.33, and recursive ReFeX values including neighbor-degree sum 17 and second-iteration sum 69. section 10.1.3 section 10.3.1
Table 10.10 Details of the egonet structure for the nodes in figure 10.9
|
Node’s egonet
|
# of nodes
|
# of internal edges
|
# of in/out edges
|
|---|---|---|---|
|
A
|
7
|
7
|
9
|
|
B
|
4
|
3
|
5
|
|
C
|
2
|
1
|
1
|
|
D
|
5
|
6
|
4
|
|
E
|
3
|
3
|
3
|
|
…
|
…
|
…
|
…
|
Inspect the node-by-node table of egonet size, internal edges, and edges connecting the egonet outward. section 10.3.1
