Evaluation of Approaches for Most Frequent Sense Identification in Russian

摘要

In this paper, we compare several approaches for determining the most frequent senses of ambiguous words for Russian. We compare several approaches (frequency-based, topic models, information-retrieval and embedding-based) and consider two representation forms of information about multiword expressions described in RuThes. We found that the information-retrieval approach is better than the method based on probabilistic topic models. The best results are obtained with the application of distributional vector representations with thesaurus path weighing.