Research on non-dependent aspect-level sentiment analysis

As a popular research field of sentiment analysis, aspect-based sentiment analysis focuses on the emotional expression in different aspects. However, the current research is not precise enough in dividing the aspects of sentiment analysis. The problem of semantic overlap between aspects occurs. Furthermore, in many cases, one aspect may be contained in several sub-aspects. When the study only focused on the emotion tends in one or several sub-aspects, the results of sentiment analysis may be distorted. To deal with these problems, we propose the concept of non-dependent aspects by analyzing the dependencies among aspects and a method for dividing non-dependent aspects. Through theoretical analysis, we demonstrate that our proposed sentiment analysis results based on non-dependent aspects are more accurate than the original one, and non-dependent aspects can be easily transferred to a new corpus. The experiments on real-world data are also supporting the results of theoretical analysis. The range of accuracy of non-dependent aspects is improved by 1.9%-13.4% than before.(c) 2023 Elsevier B.V. All rights reserved.

Automated summary

As a popular research field of sentiment analysis, aspect-based sentiment analysis aims to analyze emotional expressions in different aspects. However, current research lacks precision in dividing aspects, leading to semantic overlap issues. To address these problems, we propose the concept of non-dependent aspects and a method for their division based on analyzing aspect dependencies. Theoretical analysis and real-world experiments demonstrate that sentiment analysis based on non-dependent aspects provides more accurate results compared to traditional methods.