Forecasting user perceptions of mHealth apps: AI-driven insights from large-scale user-generated content

Abstract

User perceptions of mHealth apps are critical for forecasting adoption trends, optimizing app design, and evaluating their broader societal implications for public health and digital inclusion. Understanding how users engage with these applications is essential for their sustained use. This research incorporates AI-driven methodologies to systematically analyze large-scale user-generated content (UGC), providing predictive insights into consumer behavior and digital health engagement. Through three interconnected stages, this paper contributes to technological forecasting, digital health management, and marketing analytics by applying Natural Language Processing (NLP) and Large Language Models (LLMs) to classify brand associations in mHealth app reviews. At the first stage, 849,918 reviews from the most downloaded mHealth apps in the US were analyzed and categorized into tracking, nutrition, step counters, and rest/meditation apps. Using BERT-based topic modeling (BERTopic) and KMeans clustering, we classify key topics under Keller's brand association dimensions. At a second stage, a predictive classification model was developed using fine-tuned DistilBERT. At a third stage, an ANOVA analysis was used to examine differences in user attitudes based on brand associations and app type. Findings highlight the high number of product-related attributes mentioned in user conversations. However, emotional benefits are those driving higher user satisfaction with mHealth apps.