AMAR_ABSA: Arabic Mobile App Reviews Dataset Targeting Aspect-based Sentiment Analysis Tasks

Aspect-based sentiment analysis (ABSA) is a fine-grained analysis aiming to extract an entity's aspects and corresponding sentiments from sentences. The number of publicly available datasets written in Arabic in this field is minimal and supports very few domains. One domain not well-examined in Arabic ABSA is the mobile app domain. To the authors' knowledge, there is no public dataset in the mobile app domain targeting ABSA tasks in Arabic. This study provides a dataset with statistics and a baseline evaluation that researchers can use to conduct Arabic Aspect-Based Sentiment Analysis. The dataset contains Arabic mobile app reviews collected from three well-known apps in the Middle East; 2000 reviews were manually annotated with five labels representing aspects and four values of polarities. This dataset will provide the needed publicly available data for researchers to perform various tasks, such as sentiment analysis, aspect category/sentiment classification, and rate prediction.