Inducing Domain-specific Noun Polarity Guided by Domain-independent Polarity Preferences of Adjectives

In this paper, we discuss how domain-specific noun polarity lexicons can be in-duced. We focus on the generation of good candidates and compare two ma-chine learning scenarios in order to estab-lish an approach that produces high pre-cision. Candidates are generated on the basis of polarity preferences of adjectives derived from a large domain-independent corpus. The polarity preference of a word, here an adjective, reflects the distribution of positive, negative and neutral arguments the word takes (here: its nominal head). Given a noun modified by some adjectives, a vote among the polarity preferences of these adjectives establishes a good indica-tor of the polarity of the noun. In our ex-periments with five domains, we achieved f-measure of 59 % up to 88 % on the basis of two machine learning approaches car-ried out on top of the preference votes.