Co-occurrence Networks for Word Sense Induction

Word sense induction (WSI) is the unsupervised and knowledge-free task of clustering occurrences of homonymous or ambiguous words by their meanings. This problem has been relevant since the second half of the 20th century [1] and arises in various natural language processing areas, such as machine translation, sentiment analysis, chatbots, etc. In this work, we applied graph neural networks to the WSI-problem using a co-occurrence network and evaluated it on the RUSSE'2018 task [2]. Proposed approach demonstrates satisfactory results with low consumption of computational resources.