Abstract:Large Language Models (LLMs) have recently displayed their extraordinary capabilities in language understanding. However, how to comprehensively assess the sentiment capabilities of LLMs continues to be a challenge. This paper investigates the ability of LLMs to detect and react to sentiment in text modal. As the integration of LLMs into diverse applications is on the rise, it becomes highly critical to comprehend their sensitivity to emotional tone, as it can influence the user experience and the efficacy of sentiment-driven tasks. We conduct a series of experiments to evaluate the performance of several prominent LLMs in identifying and responding appropriately to sentiments like positive, negative, and neutral emotions. The models' outputs are analyzed across various sentiment benchmarks, and their responses are compared with human evaluations. Our discoveries indicate that although LLMs show a basic sensitivity to sentiment, there are substantial variations in their accuracy and consistency, emphasizing the requirement for further enhancements in their training processes to better capture subtle emotional cues. Take an example in our findings, in some cases, the models might wrongly classify a strongly positive sentiment as neutral, or fail to recognize sarcasm or irony in the text. Such misclassifications highlight the complexity of sentiment analysis and the areas where the models need to be refined. Another aspect is that different LLMs might perform differently on the same set of data, depending on their architecture and training datasets. This variance calls for a more in-depth study of the factors that contribute to the performance differences and how they can be optimized.

Latent Discriminative Models for Social Emotion Detection with Emotional Dependency

Emotion Detection in Online Social Networks: A Multilabel Learning Approach

Dependency-Aware Attention Model for Emotion Analysis for Online News.

Affective Topic Model for Social Emotion Detection

Towards building a social emotion detection system for online news.

Reader's emotion prediction based on weighted latent Dirichlet allocation and multi-label k-nearest neighbor model

Hidden Topic-Emotion Transition Model for Multi-Level Social Emotion Detection

Predicting Social Emotions from Readers’ Perspective

Mining Social Emotions from Affective Text

Joint Emotion-Topic Modeling for Social Affective Text Mining

Multi-Label Maximum Entropy Model for Social Emotion Classification over Short Text

Emotion recognition of social media users based on deep learning

Do Large Language Models Possess Sensitive to Sentiment?

Online Behavioral Analysis with Application to Emotion State Identification

Emotional Analysis Model for Social Hot Topics of Professional Migrant Workers

Emotion Detection with Neural Personal Discrimination

A novel method for recognizing emotions of weblog sentences

Mining Emotions of the Public from Social Media for Enhancing Corporate Credit Rating

CNN-LSTM neural network model for fine-grained negative emotion computing in emergencies

LDA-TIM: An Approach for Individual Sentiment Prediction in Social Networks.