# Sentiment Analysis in the Era of Large Language Models: A Reality Check

**Authors:** Wenxuan Zhang, Yue Deng, Bing Liu, Sinno Jialin Pan, Lidong Bing  
**arXiv:** [2305.15005](https://arxiv.org/abs/2305.15005)  
**Date:** 2023-05-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.15005v1/

## Key points

- Large language models (LLMs) like ChatGPT have opened up new possibilities for sentiment analysis.
- The paper investigates the performance of LLMs in various sentiment analysis tasks.
- Tasks include conventional sentiment classification, aspect-based sentiment analysis, and multifaceted analysis of subjective texts.
- LLMs perform well in simpler tasks but struggle with more complex tasks requiring deeper understanding or structured sentiment information.
- LLMs excel in few-shot learning settings with limited annotation resources.
- Current evaluation practices have limitations in assessing LLM's sentiment analysis abilities.
- The proposed SentiEval benchmark provides a realistic assessment of LLM's performance in sentiment analysis.
- Further advancements are needed to enhance LLM's understanding and handling of complex sentiments.
- The SentiEval benchmark is a valuable resource for future research and evaluation in sentiment analysis.

## Summary

The paper titled "Sentiment Analysis in the Era of Large Language Models: A Reality Check" by Wenxuan Zhang, Yue Deng, Bing Liu, Sinno Jialin Pan, and Lidong Bing explores the capabilities of large language models (LLMs) in performing various sentiment analysis tasks. Sentiment analysis has been a prominent area of research in natural language processing as it provides valuable insights into human sentiments and opinions. The emergence of LLMs like ChatGPT has opened up new possibilities for their application in sentiment analysis. However, the extent to which existing LLMs can be effectively utilized for different sentiment analysis tasks is still uncertain. This paper aims to address this gap by conducting a comprehensive investigation into the performance of LLMs across multiple sentiment analysis tasks. These tasks include conventional sentiment classification, aspect-based sentiment analysis and multifaceted analysis of subjective texts. To evaluate the performance of LLMs, the authors compare their results against small language models (SLMs) trained on domain-specific datasets. The study encompasses 13 tasks on 26 datasets. The findings reveal that while LLMs demonstrate satisfactory performance in simpler tasks they struggle with more complex tasks that require deeper understanding or structured sentiment information. However, LLMs excel in few-shot learning settings where annotation resources are limited; suggesting their potential for practical applications when data availability is scarce. The authors also highlight the limitations of current evaluation practices in assessing LLM's sentiment analysis abilities; this benchmark aims to offer a realistic assessment of LLM's performance in sentiment analysis. In conclusion, this paper sheds light on the capabilities and limitations of large language models in performing various sentiment analysis tasks; emphasizing the need for further advancements to enhance their understanding and handling of complex sentiments. The proposed SentiEval benchmark provides a valuable resource for future research and evaluation in the field of sentiment analysis; data and code used in the study are available on GitHub for further exploration.