[Python] How to assign subgraph IDs based on weakly connected user pairs, but split when no...

I'm working with a dataset where I want to assign a sub_graph ID to user interactions. Each row in the data represents a directed edge between an actor_user_id and a related_user_id.

I want to compute a sub_graph ID such that:

Rows belong to the same sub_graph if they are connected (even indirectly) through shared users.

However, if two edges only share the same actor_user_id, they should not be grouped unless their related_user_ids are also connected.

Here’s a simplified example:

import pandas as pd
import networkx as nx
import matplotlib.pyplot as plt

# Dummy data
edges = pd.DataFrame({
'properties_id': ['A', 'A', 'A', 'A'],
'global_journey_id': ['A1', 'A1', 'A1', 'A1'],
'actor_user_id': ['abc', 'abc', 'pat', 'abc'],
'related_user_id': ['def', 'efg', 'def', 'lal'],
})


actor_user_id related_user_id sub_graph
abc def 1
pat def 1
abc efg 2
abc lal 3


I've tried using NetworkX like this:

G = nx.DiGraph()
G.add_edges_from(zip(edges['actor_user_id'], edges['related_user_id']))
components = list(nx.weakly_connected_components(G))


But this gives me only one connected component because abc is a common node across many edges.

Question:

How can I build logic that ensures subgraphs are only formed when there is a true "shared connection", not just a shared actor_user_id?

Is there a way to force the graph to disconnect branches that don’t have overlapping related_user_ids?

Any idea or trick to break down these graphs properly (even custom component grouping logic) would be super appreciated

Continue reading...