SAtour: Sentiment Analysis Of Saudi Arabia Tourism Tweets

SAtour is a new dataset of Arabic tweets in the tourism domain for Arabic SA. The total corpus size is 2293 tweets, after manual annotation these tweets were labeled as either positive, negative, or neutral. We present the detailed process of collecting, filtering and annotating the data, we also run different experiments to provide benchmark results of Arabic sentiment classification. Benchmark results on our dataset on three-way sentiment classification shows that the highest performing baseline model was MARBERT with an accuracy up to 83%, which was pre-trained for Arabic on a massive amount of data. It should be noted that this model has enhanced its performance additionally after pre-trained on a dialectical Arabic and modern standard Arabic corpus. The dataset also was utilized for zero-shot learning method to predict sentiments and compare these sentiments with the manual annotations.