Comparison of VADER and TextBlob labeling for sentiment analysis using machine learning and deep learning models: A study on generative AI user experience

Abstract

Acquiring awareness, understanding, and emotional engagement with generative artificial intelligence (AI) is crucial for enhancing service delivery and attracting new users. This study investigates user experiences with generative AI through comprehensive sentiment analysis. We analyzed 88,343 user comments using sentiment classification, topic modeling, and emotional analysis to examine user perceptions. Topic modeling revealed key themes shaping user attraction, including idea generation, learning, support, and task assistance. Trust in ChatGPT was found to be closely linked to perceived content quality and accuracy, with applications spanning education, productivity, programming, and general exploration. For sentiment labeling, we employed both VADER and TextBlob, with results showing that TextBlob-labeled datasets generally improved the performance of machine learning and deep learning classifiers. Among the models, the GRU achieved the highest accuracy of 96.38%, followed by the SVM at 94.54%. These findings suggest that while lexicon-based labeling methods, such as TextBlob, enhance classification effectiveness, the strongest predictive performance is achieved by advanced machine learning and deep learning models. Overall, the research highlights that user acceptance of ChatGPT is shaped not only by its practical utility but also by the emotional connections it fosters and its relevance across diverse contexts. These insights provide valuable perspectives on the dynamics of human-AI interaction.

Ähnliche Arbeiten