Multi-senses and multi-dependencies discovery among words

Word sense and word dependency benefit many applications. Manually constructed lexicon usually serves as a source for word sense and word dependency. However, there always requires very expensive work and extensive time consumption to compile it, simultaneously senses and relationships among words in these kinds of lexica are always missed. Automatically compiled lexica are under developing, the main obstacle is to discover multisenses and multidependencies among words. We propose a new method combining an improved ISODATA clustering algorithm with association rule mining to answer the question. With the recursively clustering algorithm lower frequency senses are discovered. As well a approach for refinement is put forward to improve the precision. Experiments indicate that the approach presented here provides preferable outputs.