LAUSR

Comments from social media platforms (such as YouTube) have become a valuable resource for manufacturers to examine public opinion toward their products. Accordingly, we propose a novel framework for automatically collecting, filtering, and analyzing comments from YouTube for a given product. First, we devise a classification scheme to select relevant and high-quality comments from retrieval results. These comments are then analyzed in a sentiment analysis, where we introduce a joint approach to perform a combined sentence and aspect level sentiment analysis. Hence, we can achieve the following: (1) capture the mutual benefits between these two tasks, and (2) leverage knowledge learned from solving one task to solve another. Experiment results on our dataset show that the joint model achieves a satisfactory performance and outperforms the separate one on both sentence and aspect levels. Our framework does not require feature engineering efforts or external linguistic resources; therefore, it can be adapted for many languages without difficulties.