A survey on review summarization and sentiment classification

Abstract

With increasingly more people using online services, purchasing products, and reviewing them, it becomes crucial to have a system that can provide a crisp representation of thousands of reviews written by them. This representation must depict the sentiment that a user has toward the product. This notion comes under review summarization (RS). An even more concise and generous representation to a customer would be a label: positive, negative, or neutral, depicting the reviewer’s opinion toward the product/service. This comes under the domain of sentiment classification (SC). There have been several advancements in both RS and review SC techniques through the years. Some very recent techniques have tried to perform these two tasks jointly, and their results have depicted that these tasks can, in fact, mutually benefit one another in improving their performance. This paper presents contemporary and some earlier techniques used in both RS and SC and the more recent techniques where both these tasks are performed jointly. We have also performed experiments on joint models and devised a model with a combination of deep learning, rule-based systems, and evolutionary algorithms. © 2022, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.