MABSA: Malayalam Aspect Based Sentiment Analysis Dataset

MABSA is a curated aspect-based sentiment analysis dataset on the South Indian language Malayalam over the movie review domain. This dataset contains 4000 movie reviews in Malayalam and its sentiment labels for nine target aspects: direction, acting, songs, screenplay, story, casting, editing, cinematography, and BGM (background music). Each aspect in the review is mapped to one of the three sentiment classes, which are positive, negative, or neutral. These three sentiment classes are numerically represented as 1, 0, and 0.5, respectively. Movie reviews in Malayalam are collected through two methods, where the first method uses the Google survey tool to collect reviews from people, and the second method manually gathers movie reviews from various pages and groups on different social media platforms like Facebook, YouTube, and IMDb. The MABSA dataset consists of 4000 Malayalam movie reviews and a total of 5707 aspects manually annotated with its sentiment orientation. Dataset files include both with and without English translations of Malayalam movie reviews. A copy of the Google form file is attached along with the MABSA dataset.