A Survey of Text Aspect-level Sentiment Classification Focusing on Relevance

Sentiment classification is an essential part of sentiment analysis. In recent years, it has emerged as a major research area in natural language processing. This field of research has high application value in social media, product reviews, and other areas. With the advent of deep learning, research for sentiment classification is evolving from coarse-grained towards fine-grained approaches. The merge is relatively oriented towards aspect-level sentiment classification. The underlying technologies for aspect-level sentiment classification consist mainly of neural networks and their hybrid models. The text level is the focus of this study, which highlights the relationships between texts. It is a classification of the kinds and methods of text sentiment categorization. It presents a comparative study of currently-available aspect-level sentiment classification techniques. It details the structural features, application mechanisms of different models and quantitative evaluation of experimental data. This yields to an assessment of the pros and cons of various models. Moreover, this study proposes the framework of "aspect relevance.", which gives qualitative evaluation of how well various methods solve textual relations. This study provides some outlooks for research directions on aspect-level sentiment classification in the future. It indicates future models should find the right equilibrium between performance and cost. This study also underlines the need to strengthen generalization abilities and calls for multimodal data representations and new application domains.