Using distributional knowledge for sentiment analysis

Abstract

Conventional approaches for Sentiment Analysis are either based on Sentiment Lexicons or Classification algorithms accompanied with feature engineering. A major drawback of these approaches is vocabulary dependency. On the other hand, knowledge acquisition in Distributional Semantics depends either on la.tent or on explicitly defined concepts from vast knowledge bases. Linguis tic entities like words, phrases, sentences a.re represented as multidimensional vectors of concepts. Therefore, if we employ Distributional Semantics, a word occurring in a new text need not be available in a sentiment lexicon or a part of a training vocabulary; it could be identified by its underlying concepts. Distributional Semantics is finding its way to the current changes in NaturaI Language Processing and has been proved to be very useful in areas like context disambiguation, word sense disambiguation, machine translation etc