“…Afterwards, one of our internal researchers manually classified each keyword into one of the three categories if applicable. As for the rest, nouns and meaningless words like 'is' or 'not' were removed and its parts of speech was checked using WordNet from Python NLTK package 4 . For all 8 paintings, the object keywords were the most frequent (23.2%), followed by hue, location, status, other, shape, size, brightness, texture(18.0%, 14.3%, 10.4%, 10.0%, 8.3%, 4.8%, 4.2%, 3.4%, respectively.…”