Remove the commented code and provide a description in README.md

10-disambiguate.py

@@ -45,7 +45,6 @@ def emit(word):
             candidates   = [(nsid, sim(vector, v.transform(neighbours[nsid]))[0, 0]) for nsid in neighbours]
             nsid, cosine = max(candidates, key=itemgetter(1))
             if cosine > 0:
-            # if cosine > 0 or len(candidates) == 1:
                 nsense = '%s#%d' % (neighbour, nsid)
                 sneighbours[sense][nsense] = weight
 
@@ -56,20 +55,3 @@ def emit(word):
         for sense, neighbours in sneighbours.items():
             for nsense, weight in neighbours.items():
                 print('%s\t%s\t%f' % (sense, nsense, weight))
-
-# for word, senses in wsi.items():
-#     for sid, words in senses.items():
-#         sense  = '%s#%d' % (word, sid)
-#         vector = v.transform(words)
-#         sneighbours = {}
-
-#         for neighbour, weight in words.items():
-#             neighbours = wsi[neighbour]
-#             items = [(nsid, sim(vector, v.transform(neighbours[nsid]))[0, 0]) for nsid in neighbours]
-#             nsid, cosine = max(items, key=itemgetter(1))
-#             if cosine > 0:
-#                 nsense = '%s#%d' % (neighbour, nsid)
-#                 sneighbours[nsense] = weight
-
-#         for nsense, weight in sneighbours.items():
-#             print('%s\t%s\t%f' % (sense, nsense, weight))

README.md

@@ -1,2 +1,9 @@
-# concept-discovery
-Concept Discovery from Synonymy Graphs
+# Concept Discovery from Synonymy Graphs
+
+This is an implementation of concept discovery (or synset induction, if you wish) approach that uses synonymy graphs. It is based on three algorithms:
+
+* ego-network clustering for inducing the word senses ([Panchenko et al., 2016](https://www.linguistics.rub.de/konvens16/pub/24_konvensproc.pdf)),
+* cosine-based sense disambiguation for disambiguating them ([Faralli et al., 2016](http://link.springer.com/chapter/10.1007/978-3-319-46547-0_7)),
+* Chinese Whispers for graph clustering ([Biemann, 2006](http://dl.acm.org/citation.cfm?id=1654774)).
+
+However, it is not based on distributional methods.