Aspect Based Sentiment Analysis of Product Review Using Memory Network

Abstract. Purpose: Consumer opinion is one of the essential keys that affect success a product. Sentiment analysis consumer needed to find out information about customer satisfaction for companies in decision-making process. The traditional sentiment process extracts complete from single sentence. However, it does not consist only total number depends on aspects make up Therefore, pay attention aspects.Methods: This research focuses product reviews Indonesian e-commerce several sentiment. Uses fastText word embedding avoid Out Vocabulary datasets and Gated Recurrent Units aspect spread detection. classification using Memory Network method.Result: experiment results showed aspect-based predictions had an accuracy 83% compared 78% overall review texts, indicating can improve model performance predictions.Novelty: Most use document-level extract predict reviews, be applied better understanding, store important explicitly polarity.