Co-occurrence metrics

This method is often used to find related tags, or to cluster the tags.

The co-occurrence coefficient between two tags is the number of resources (photos, links etc.) where both tags are assigned to. To account for the individual frequencies, the co-occurrence value can be normalized.

Symmetric normalization can be done using the Jaccard coefficient.

Asymmetric normalization can be done by

${{|t_{i}\cap t_{j}|} \over {|t_{i}|}}$

where $|t_{i}\cap t_{j}|$ is the number of times tags $t_{i}$ and $t_{j}$ occur together and $|t_{i}|$ is the number of times tag $t_{i}$ occurs.