Evaluation method for Neo4j recommendation

Dear friends, I'm trying to create a real-time recommendation system using the Neo4j graph database. And I want to write a paper base on that.

I'm using different CQL to find the best matches to suggest as recommendations to users through algorithms like cosine similarity.

There is a problem that I'm struggling with, I don't know how to evaluate my system or what should I compare my recommendation system with to prove my solution is good (an acceptable solution which is also accepted by academia).

The big problem is since all my data are in Neo4j I think mean absolute error and mean squared error (MAE and RMSE ) are not suitable for my situation! I mean how can I do that with data in neo4j? is there a better solution here for evaluation or if it is not, is there any example for calculating MAE and RMSE in Neo4j?

PS: this is a real-time and graph-based recommendation solution.