Unveiling user challenges in immersive technologies: A deep learning approach to social media analytics

Abstract

Organizations are increasingly exploring the integration of immersive technologies into their business models. Understanding the barriers to adoption from the user's perspective is essential for successful implementation. This study proposes a deep learning framework to analyze social media data and uncover user-reported challenges associated with immersive technologies, including virtual reality (VR), augmented reality (AR), and mixed reality (MR). To detect adverse user experiences, we employ a semi-supervised learning approach based on Bidirectional Encoder Representations from Transformers (BERT), a context-aware language model developed by Google in 2018 and widely used in natural language processing. This approach incrementally builds a sentiment prediction model to identify negative user posts. We then apply BERTopic, a topic modeling technique built upon BERT, to classify these posts into semantically coherent topics. Finally, the identified topics are evaluated based on post volume, growth rate, and their strategic positioning using a topic strategy map. The analysis reveals 21 topics for VR, 8 for AR, and 8 for MR. These reflect a wide spectrum of concerns, including hardware malfunctions, tracking instability, content limitations, user discomfort, and governance skepticism. While some issues are shared across modalities, others, such as controller mapping failures in MR or WebAR instability in AR, are uniquely emphasized. The findings offer practical insights to guide user-centered design, improve platform reliability, and support the broader adoption of immersive technologies.