Automatic aspect identification and clustering are critical tasks for opinion mining/sentiment analysis, as users employ varied terms (explicitly or not) to evaluate objects of interest and their characteristics. In this paper, we focus on aspect clustering methods and present a new approach to group implicit and explicit aspects from online reviews. We evaluate four linguistic methods inspired in the literature and one statistical method (using word embeddings), and also propose a new one, based on varied linguistic knowledge. We test the methods in three commonly used domains and show that the method that we propose significantly outperforms the other methods by a large margin.
展开▼
