Author_Institution :
Dept. of Inf. & Comput. Sci., Keio Univ., Yokohama, Japan
Abstract :
A system able to interpret coherence of multiple sequential foci, while disambiguating multiple meanings of representations arising in the context, is exposed in this article. In the e.g., [I disliked in my childhood the bitterness of green peppers, it tastes delicious now.], identifying which of ldquodislikerdquo or ldquodeliciousrdquo is the main meaning, evaluation, among these sentences is necessary. Also detecting the antecedent of ldquoitrdquo, an anaphor, is as well indispensable for interpreting accurately the knowledge representation expressed in this contextual sentence. In the following example, having a ldquobig heartrdquo (a tolerant mind), ldquobigrdquo is usually understood as a quality, whereas in the expression ldquobig wordsrdquo, it has rather a pejorative connotation. To resolve such ambiguity, this system achieves an accurate contextual analysis, by identifying compositionally the overall role relations between words. For the determination of the evaluation polarity, in a corpus composed of 2500 customer reviews, the correct answer rate is 75.0%.
Keywords :
knowledge representation; natural language processing; anaphor detection; contextual coherence interpretation; contextual sentence analysis; customer review corpus; evaluation polarity; knowledge representation; multiple meaning disambiguation; multiple sequential foci; natural language processing method; pejorative connotation; Brightness; Computer science; Data mining; Electronic mail; Heart; Internet; Knowledge representation; Mood; Natural languages; Object detection;