SoftAttributes

홈페이지 · 논문 4편

The dataset consists of sets of movie titles, with each set annotated with a single English soft attribute (subjective descriptive property, such as 'confusing' or 'romantic') and a reference movie. For each set, a crowd worker has placed the movies into three sets: more, equally, and less than the reference movie. There are 5,991 such sets, from which one can infer approximately 250,000 pairwise preferences over movies for the 60 distinct soft attributes studied.