An Empirical Study on Sentiment Intensity Analysis via Reading Comprehension Models

Sentiment intensity analysis, a significant branch of natural language processing (NLP), has garnered widespread attention due to its potential to extract valuable insights from user-generated content on various internet platforms. This information can aid individuals and consumers in making informed decisions. Applications of this technology span multiple domains, including public opinion analysis, where it monitors and controls sentiment trends, and stock market analysis, which often reflects economic conditions and capital flows. Current methods for sentiment intensity analysis encompass lexicon-based methods, traditional machine learning techniques, deep learning models, and large language models. With the rapid advancement of deep learning technologies and the availability of extensive digital datasets, sentiment intensity analysis has achieved a relatively mature state. However, there has been little work comparing how reading comprehension models perform on sentiment intensity analysis tasks. In this paper, we explore the implementation of sentiment intensity analysis through reading comprehension modeling. Specifically, we experiment with five models, including pretrained, extractive, and generative models. We conduct extensive experiments on two distinct sentiment intensity analysis datasets: movie scripts and restaurant reviews. Our findings shed light on the feasibility and adaptability of reading comprehension models for sentiment analysis tasks, offering valuable insights for the continued development and application of sentiment intensity analysis.