Multi-class bootstrapping learning aspect-related terms for aspect identification

Aspect identification in entity reviews involving multiple aspects is a top priority for aspect-based opinion mining. Most of previous studies adopted machine learning techniques taking it as a multi-class text classification task. However, since building labeled training data is often expensive, some researchers put more interest in unsupervised techniques. With the subject of online restaurant reviews, this paper presents a new multi-class bootstrapping algorithm to learn aspect-related terms to be used for aspect identification. Experimental results demonstrate that our method without requiring labeled training data achieves good performance in comparison to the state-of-the-art supervised learning techniques.