/bcubed_f1

Implementation of the BCubed F1 score for measuring similarity of two community detection partitions

Primary LanguagePython

BCubed F1

Implementation of the BCubed F1 score for measuring similarity of two community detection partitions.

The BCubed measure was originally proposed to evaluate effectiveness of document clustering paper. Its properties were compared to a wide range of other extrinsic clustering evaluation metrics, with the conclusion that BCubed satisfies all the required qualitative properties. Since data clustering and community detection in networks produce analogous results, one can also apply the BCubed measure to evaluate the detected communities.

The F1 is calculated node-wise as a sum of the Precision and Recall of individual nodes. As the Precision and Recall are normalized by the number of nodes in the network, the F1 ranges from 0 (no similarity between the partitions) to 1 (complete match between the partitions). A detailed definition of the Bcubed F1 metric for communities, including the general form for non-identical node sets, can be found in this work.