Abstract:

This study explores the possibility of using emojis as a feature for distant supervision annotation of a large-scale dataset for sentiment analysis. In the process we propose a newly collected Arabic dataset for sentiment analysis with over than 2 million labeled tweets. The data was gathered from Titter during the period between the 1st of June and the 30th of November 2017. We also present a new deep learning model for Arabic sentiment analysis with bidirectional gated recurrent units. Our model achieved state of the art performance with 88% of accuracy.