Does emotion influence the use of auto-suggest during smartphone typing?
summary
Built an Android keyboard logging app and runs a three-week in-the-wild study with ~20 installs, 13 users, ~3,000 typing sessions; labels sessions as accepted if any suggestion is picked (28% accepted) and collects self-reported emotion (happy, sad, stressed, relaxed). Trains personalized random-forest models with three emotion features (current, previous, elapsed), outperforms baselines (baseline 73% → balanced via SMOTE to AUCROC 82%), and finds higher suggestion acceptance in low-activity states (sad, relaxed).
AI-written summary · about