Abstract
With the spread of social media, the demand for automated systems that analyze these massive amounts of data on the Web is increasing. One domain for these systems is sentiment analysis(SA). SA is designed to extract sentiment from text; this is often accomplished by using lexicons that indicate the sentiment polarity of words. While there are many English lexicons that are available, there is a lack of Arabic lexicons. In previous work, an attempt was made to generate an Arabic sentiment lexicon extracted from Twitter using the Pointwise Mutual Information (PMI) statistical method. In this paper, we extend the work by using two different statistical approaches: Chi-Square and Entropy to generate the lexicons. Intrinsic and extrinsic evaluation was conducted to compare the three lexicons. The results showed the superiority of PMI.