Best way to create Nodes and Relation for a given data where a node is linked to multiple nodes

Hi everyone,

I want to know about the best way to construct a knowledge graph based on the give data in the table.
Description: The same player 'John' plays for two clubs 'Club A' and 'Club B', For Club A, he scored a total of 5 goals while for Club B he scored 3 goals.

How to represent this in a Graph specifically Neo4J? (Keeping in mind that I have to use those relation triplets later for recommendation based on Graph embedding)


1- From the above tables, the following graph 1 shows the simplest graph but it is ambiguous for John to which club he scored 5 goals and for which 3 goals.

2- Graph 2 is pretty straight forward but in that case we have to create two separate nodes for the same entity to distinguish the scored property for the particular club.

3- Graph 3 is similar to that of 1 while it adds and additional relation between club node and score node with a predicate as JOHN_SCORED

These are the possible ways I could think of creating graph. Please guide me which approach is better or is there another better approach.

** After generating graph, I have to apply Deep learning with Graph embeddings and GNNs.


Hi engr.muzamilshah!

I'd suggest a schema closest to Graph 3 where you have player, team, and goal nodes, the number of goals that player scores being a property of that goal node. In this way your schema also becomes more scalable should you want to add additional game statistics. Hope this helps!


1 Like
Here is my solution:

MERGE (a:Player {name: "John"})
MERGE (b:Club {club: "A"})
MERGE (c:Club {club: "B"})

MERGE (d:Score {score:5})
MERGE (f:Score {score:3})

MERGE (a)-[:PLAYS_FOR]->(b)
MERGE (b)-[:SCORE]->(d)

MERGE (a)-[:PLAYS_FOR]->(c)
MERGE (c)-[:SCORE]->(f)

return a, b, c, d, f


1 Like

The "goals scored" could also sit as a property of the relationship "played for", quantifying the relationship.

Lot depends on how you intend to query it, quantifying the relationship may be slightly less performant, but I can't imagine a sport where there is enough data to matter, and it seems a very natural place to put it.

1 Like

Hi @rebeccarabb , @ameyasoft and @waters.simon

Thank you all for valuable hints and suggestions.
With all these comments, I clearly got the idea about how should I carry on the graph construction process depending on my use case.

Thanks again :slight_smile:

1 Like