Metric for label propagation algorithm

I ran label propagation algorithm on my data and I would like to know if there is a metric I can use to evaluate the model.

Is there a metric for this algorithm ? If no what can I do to evaluate this algorithm ?

This algorithm is a community detection algorithm.

What metric are you looking for, I must imagine it relates to your specific use case, since this is just an algorithm which (assuming property implemented) just gives it's output results.

If you have some form of ground truth community information for your dataset and want to evaluate if label propagation is a reasonable community detection algorithm to use with your dataset you could create a metric around that information using the true/false positives, and true/false negative information, in the form of AUC ROC analysis.

This is discussed in the Neo4j Graph Algorithms book and there is also a lot of information online about AUC ROC if you google it.

Connecting with duplicate for reference: https://stackoverflow.com/questions/63520403/neo4j-metric-for-label-propagation-algorithm-lpa

1 Like

You can check the didConverge parameter to see if LPA yielded a stable result -- if it's false, then you need to set your number of iterations higher.

1 Like

Thank you @Joel for your answer. My project is to make clusters of articles having similar subjects. So I look for something that enables me to know if the clusters are good.

Thank you @alicia.frame