Analysis of emotional attitudes toward autonomous vehicles: A comparative approach using language models

Abstract

Autonomous vehicle technologies are rapidly developing worldwide, bringing various societal and individual viewpoints to the forefront. With the widespread use of social media platforms, the number of user-generated comments about this technology has significantly increased. Thanks to advances in natural language processing (NLP), the analysis of such text-based data can now be conducted in a more meaningful and systematic way. Accurate analysis of users' emotional attitudes toward the technology can provide strategically valuable insights for companies and policymakers. This study aims to identify users' emotional states regarding autonomous vehicle technology. To this end, a total of 48,897 English-language comments were collected from the platforms Quora, YouTube, and Reddit and analyzed. The study consists of three main stages: In the first stage, four different pretrained language models (BERT, RoBERTa, DeBERTaV3, and XLM-RoBERTa) were fine-tuned using the dair-ai/emotion dataset to classify six basic emotion categories. In the second stage, a comprehensive text dataset was created from unlabeled user comments collected from social media platforms. In the third stage, the RoBERTa model, which achieved the highest accuracy in performance comparisons, was used to categorize these comments into emotion categories. The findings revealed that the majority of users expressed joy toward autonomous vehicles, while a significant portion also exhibited negative emotions such as anger and fear. These results indicate that emotional responses can serve as a guide in the societal acceptance process of autonomous vehicles.