Sinhala Sentiment Analysis using Corpus based Sentiment Lexicon

Sentiment analysis has become a much needed natural language processing technique as the usages of sentiment analysis based tools are getting popular. With sufficient amount of lexical resources available, an accurate sentiment analysis can be performed for English. Sinhala is a morphologically rich but a less resourced language. Availability of lexical resources directly effect on the success of the development of natural language processing tools as well as for further researches. Sentiment lexicon is a primary resource required to perform a lexicon based sentiment analysis in any language. We propose a semi-automated framework for constructing a sentiment lexicon for Sinhala Language using a corpus based method. Proposed framework used conjunction as the indicator of polarity for adjectives in a conjoined adjective pair and morphological relationships between adjectives to determine the polarity of adjectives. To the best of our knowledge, such method has not been attempted for sentiment lexicon construction in Sinhala, until this day. The approach was tested for a 3 million POS tagged word corpus and the resultant sentiment lexicon was used to perform a lexicon based sentiment analysis, which produced 65%-69% of accuracy for three different classifiers. Proposed framework demonstrated potential to be more accurate with sufficiently larger text corpuses.