Unsupervised aspect-based sentiment analysis on Indonesian restaurant reviews

We address the problem of aspect-based sentiment analysis (ABSA) from Indonesian restaurant reviews due to its usefulness for customer satisfaction applications. The main task is divided into two subtasks: (1) aspect extraction, and (2) aspect sentiment orientation classification. For both subtasks, we propose an unsupervised approach which does not rely too much on hard-core natural language processing tools since Indonesian language is still under-resourced in terms of language technology. Our unsupervised approach employs several techniques in the area of distributional semantics, such as word embeddings and point-wise mutual information.