Mapping <scp>UX</scp> Friction: A Neural Network Approach to Analyzing Real Time Navigation Sentiment

ABSTRACT In the context of emotional cartography, this study introduces a neural network approach to classify user sentiment from speech transcriptions collected via Think Aloud protocols. While traditional metrics capture what users do, verbal protocols reveal how they feel and think during navigation. However, analyzing such qualitative data at scale remains a bottleneck. To address this, we trained a neural network using a novel combination of a curated, domain‐specific dataset ( n = 3052) and large‐scale app store reviews. A field experiment was conducted to empirically validate the model's ability to classify immediate, in situ emotions. The results demonstrate that supplementing generic training data with domain‐specific examples significantly improves accuracy up to a saturation point. Furthermore, by spatializing these sentiments, the study maps “UX Friction” identifying specific geographic coordinates at which interface challenges induce cognitive load. This work validates a scalable, low‐cost computational pipeline for evaluating digital maps, bridging the gap between qualitative user insights and quantitative spatial analysis.