Vibe Check: A Hybrid MultiLabel Emotion and RuleBased NLP System for SubtextAware Professional Communication Analysis

Professional communication often carries nuanced emotional subtext that is not captured by binary sentiment tools. This paper presents Vibe Check, a hybrid natural language processing system combining a fine-tuned BERT multi-label emotion classifier with an interpretable rule layer to detect communication intent categories such as High Stress/Urgent, Sarcasm/Backhanded tone, and Passive-Aggressive expression. The model is trained on the GoEmotions simplified taxonomy (28 emotion labels) using weighted binary cross-entropy to address class imbalance. We report training and validation trends from project checkpoints, showing a reduction in validation loss from 0.2122 to 0.1881 and an increase in F1 from 0.4414 to 0.4944 between epochs 1 and 2. The deployed implementation includes a Flask inference API and Angular interface for real-time analysis, recommendation generation, and persistent interaction history. Results demonstrate that combining deep contextual emotion modeling with explicit decision logic improves actionability for end users while preserving interpretability.