This is a demo of the AdaGram (Bartunov et al., 2015) method, and a regular polysemy pattern explorer.

The model was built using AdaGram.jl with alpha = 0.10, window size 5, vector size 300 and maximum number of senses 10 on a 2 billion word corpus of Russian lemmatized with MyStem3. The corpus consists RuWac, Russian wikipedia and lib.ru archives.

AdaGram sense induction demo: enter any Russian word into the search field, and see induced senses, sense neighours and context words. For example: яблоко.

Regular polysemy pattern demo: enter a word and its sense numbers (denoted with a # prefix in each sense card), and see what words have similar polysemy, for example: блеять.