Closed dustalov closed 5 years ago
This does not seem to be an overlapping community detection procedure.
Only at a first glance. Watset was originally designed for the synset induction problem, which implies clustering of a synonymy graph. Each resulting cluster, called a synset, contains words with the same meaning. Of course, these clusters overlap in case of polysemous words.
Look at Figure 2 in the original Watset paper. The algorithm transforms the input graph into a sense-aware representation and then clusters it (colors correspond to the clusters). Removal of these sense labels produces overlapping clusters. In Table 3, we compare directly to very similar community detection approaches, such as Chinese Whispers, MCL and Clique Percolation Method.
We found this algorithm to be useful in other scenarios, such as semantic frame induction. You can try our Java implementation to confirm my description: https://github.com/nlpub/watset-java, it is very simple to use both with CLI and API.
This pull request adds one more overlapping community detection algorithm, Watset, to the “Other” category.