Deep Learning for Aspect-Based Sentiment Analysis

Sentiment analysis is an important task in natural language understanding and has a wide range of real-world applications. The typical sentiment analysis focus on predicting the positive or negative polarity of the given sentence(s). This task works in the setting that the given text has only one aspect and polarity. A more general and complicated task would be to predict the aspects mentioned in a sentence and the sentiments associated with each one of them. This generalized task is called aspect-based sentiment analysis (ABSA). In the annual SemEval competition, an ABSA task has been added since 2014. Among submissions of the past two years, most winning models use support vector machines (SVM). Riding on the recent trends of deep learning, this work applies deep neural nets to solve this task. We design a combined model with aspect prediction and sentiment prediction. For both predictions, we achieve better than or close to state-of-theart performance using deep learning models. We also propose a new method to combine the syntactic structure and convolutional neural nets to directly match aspects and corresponding polarities.