PerSent 2.0: Persian Sentiment Lexicon Enriched with Domain-Specific Words

Abstract

Sentiment analysis is probably the most actively growing area of natural language processing nowadays, which leverages huge amount of user-contributed data on Internet to improve income of businesses and quality of life of consumer. The majority of existent sentiment-analysis systems is focused on English, due to lack of resources and tools for other languages. To fill this gap for Persian language, in our previous work we have compiled the first version of PerSent Persian sentiment lexicon, which was small and included only words and phrases from general domain. In this paper, we present its extension with words from three different domains and evaluate its performance on polarity classification task using various machine learning-based classifiers. We use a multi-domain dataset to evaluate the performance of our new lexicon on various domains. Our results demonstrate usefulness of the new lexicon for analysis of product and movie reviews and especially of political news in Persian language.