Understanding Social Perceptions,Interactions,and Safety Aspects of Sidewalk Delivery Robots Using Sentiment Analysis

Abstract

As sidewalk delivery robots (SDRs) become increasingly common in urban environments, understanding public perception is crucial for their successful integration. This study presents an analysis of public sentiment toward SDRs by examining YouTube comments. We manually annotated the collected comments with three sentiment labels: negative, neutral, and positive, and developed machine learning models for sentiment classification (SC). These classification models enable scalable and real-time inference of public attitudes, which is crucial for monitoring societal acceptance of emerging technologies. Our results show that, in binary classification tasks, a support vector machine (SVM) with term frequency–inverse document frequency (TF-IDF) features achieved the highest accuracy, while in ternary classification tasks, a deep learning model combining bidirectional encoder representations from transformers (BERT), long short-term memory networks (LSTM), and gated recurrent units (GRU) outperformed other models, reaching an accuracy of 0.78. To further explore the public’s underlying concerns, we applied latent Dirichlet allocation (LDA) to extract ten key topics from the comments. Building on these findings, we offer targeted policy recommendations for improving the deployment, interaction, and safety of SDRs. This work provides insights for policymakers and stakeholders seeking to understand and shape public acceptance of autonomous delivery technologies.