Automatic Expandable Large-Scale Sentiment Lexicon of Modern Standard Arabic and Colloquial

Abstract

In subjectivity and sentiment analysis (SSA), there are two main requirements are necessary to improve sentiment analysis effectively in any language and genres, first, high coverage sentiment lexicon - where entries are tagged with semantic orientation (positive, negative and neutral) - second, tagged corpora to train the sentiment classifier. Much of research has been conducted in this area during the last decade, but the need of building these resources is still ongoing, especially for morphologically-Rich language (MRL) such as Arabic. In this paper, we present an automatic expandable wide coverage polarity lexicon of Arabic sentiment words, this lexical resource explicitly devised for supporting Arabic sentiment classification and opinion mining applications. The lexicon is built using a seed of gold-standard Arabic sentiment words which are manually collected and annotated with semantic orientation (positive or negative), and automatically expanded with sentiment orientation detection of the new sentiment words by exploiting some lexical information such as part-of-speech (POS) tags and using synset aggregation techniques from free online Arabic lexicons, thesauruses. We report efforts to expand a manually-built our polarity lexicon using different types of data. Finally, we used various tagged data to evaluate the coverage and quality of our polarity lexicon, moreover, to evaluate the lexicon expansion and its effects on the sentiment analysis accuracy. Our data focus on modern standard Arabic (MSA) and Egyptian dialectal Arabic tweets and Arabic microblogs (hotel reservation, product reviews, and TV program comments).