IMDB Sentiment Classification https://github.com/weakrules/Denoise-multi-weak-sources/tree/master/rules-noisy-labels/IMDB # Labels "0": "Negative", "1": "Positive" # Labeling functions lfs = [ expression_nexttime, keyword_compare, keyword_general, keyword_finish, keyword_plot ] # lf - expression_nexttime expression_nexttime = make_expression_lf(name="expression_nexttime", pre_pos=["will ", " ll ", "would ", " d ", "can t wait to "], expression=[" next time", " again", " rewatch", " anymore", " rewind"]) # lf - keyword_compare keyword_compare = make_keyword_lf(name="keyword_compare", keywords_pos=[], keywords_neg=[" than this", " than the film", " than the movie"]) # lf - keyword_general keyword_general = make_keyword_lf(name="keyword_general", keywords_pos=["masterpiece", "outstanding", "perfect", "great", "good", "nice", "best", "excellent", "worthy", "awesome", "enjoy", "positive", "pleasant", "wonderful", "amazing", "superb", "fantastic", "marvellous", "fabulous"], keywords_neg=["bad", "worst", "horrible", "awful", "terrible", "crap", "shit", "garbage", "rubbish", "waste"]) # lf - keyword_finish keyword_finish = make_keyword_lf(name="keyword_finish", keywords_pos=[], keywords_neg=["fast forward", "n t finish"]) # lf - keyword_plot keyword_plot = make_keyword_lf(name="keyword_plot", keywords_pos=["well written", "absorbing", "attractive", "innovative", "instructive", "interesting", "touching", "moving"], keywords_neg=["to sleep", "fell asleep", "boring", "dull", "plain"])