Enhancing Mobile App Ranking through Trust-Based Sentiment Analysis using Large Language Models

Abstract Selecting the “most suitable” app from a categoryremains a challenge due to the presence of multiple highly ratedalternatives offering similar functionality. Users often rely ontrial and error, including installing and testing several apps,before finding one that meets their needs. In this study, weextend previous research conducted by our group and propose anenhanced trust-based app ranking approach that takes advantageof user feedback available in the “Relevant” section of GooglePlay reviews. We collect apps from a specific category and analyzeuser reviews using three different large language models (LLMs)to extract sentiment data. From these sentiment scores, we derivea BDU tuple, representing Belief, Disbelief, and Uncertaintyabout the app. We use the BDU tuple to compute a trustscore that reflects users’ perceived trust about each app. Appsare then ranked based on these trust scores. We compare ourranking with existing app ratings, available in Google Play, usingthe Kendall Tau distance. Our approach offers a user-centricalternative to traditional rating systems by emphasizing perceivedtrustworthiness derived from real user experiences.