Business Context
A marketing firm analyzes a large social network. The initial communities are too large to be actionable for targeted campaigns. They need to recursively break down these giant components into smaller, hyper-segmented groups.
Data Preparation
Simulation of a larger graph with a dense core using a loop to generate connections.
Copied!
| 1 | |
| 2 | DATA mycas.social_graph; |
| 3 | DO i=1 to 100; |
| 4 | from_user=cat('User_', i); |
| 5 | DO j=1 to 3; |
| 6 | to_user=cat('User_', floor(100*rand('uniform')+1)); |
| 7 | IF from_user ne to_user THEN OUTPUT; |
| 8 | END; |
| 9 | END; |
| 10 | |
| 11 | RUN; |
| 12 |
Étapes de réalisation
1
Execute Recursive Community Detection with Label Propagation.
Copied!
| 1 | PROC CAS; |
| 2 | network.community / |
| 3 | algorithm="LABELPROPAGATION" |
| 4 | links={name="social_graph"} |
| 5 | recursive={maxCommSize=10} |
| 6 | outNodes={name="micro_segments", replace=true}; |
| 7 | QUIT; |
Expected Result
The action should run iteratively. The final `micro_segments` table should contain community IDs where no single community exceeds 10 nodes (maxCommSize), effectively breaking down the main connected component.