Opinion sentence classification in Indonesian review document

Opinion sentence classification in Indonesian review document has never been done yet, considering special characteristics of review document that we found in this research. Among others are the sentence type in the review document, complex sentence structure, non standard sentence pattern, non formal vocabulary, excessive punctuation usage and many mixing language. The aim of this research is to provide necessary process for opinion sentence classification in an Indonesia language review document regardless of only few low accuracy NLP (Natural Language Processing) tools available for Indonesian language. Opinion sentence classification in this research is done with sentiment and subject detection approach in sentence. By locating sentiment word in sentence and determining whether the clause subject is considered as the frequent subject which is often subjected to an opinion, one could determine whether the sentence is an opinion expression or not. The proposed opinion sentence classification require two processes: the subject accumulation to accumulate relevant subject words automatically; and sentiment classification to classify sentence as opinion sentence or not. Accuracy of the experiment (using precision score) of the method proposed on the opinion sentence classification is up to 87%.