A System-Level Framework for Sentiment-Aware Reflective Writing Systems: Modeling Temporal Emotional Patterns with Interpretability and Ethical Safety

Sentiment analysis has become a cornerstone of affective computing applications, enabling systems to detect emotional valence in usergenerated text. However, when deployed in reflective writing and coaching contexts, sentiment-based systems face a fundamental limitation: sentiment polarity does not reliably indicate user intent, therapeutic need, or appropriate system response. This paper formally analyzes the gap between sentiment detection and intent interpretation in emotion-aware interactive systems. We define sentiment as a function S : X → Y s mapping text to polarity labels, and intent as a context-dependent function I : X × C × G × H → A incorporating conversational context, user goals, and interaction history. Through system-level observations from early-stage prototype deployment, we demonstrate that S(x) alone is statistically insufficient for determining appropriate system actions. We present a multi-layer intent modeling framework that augments sentiment detection with pragmatic reasoning, temporal pattern recognition, and goal-aware interpretation. Our central contribution is a utility-theoretic framing of emotional response selection that explicitly models asymmetric costs (false intervention vs. missed support), derives optimal decision thresholds, and incorporates ethical constraints through action-space restriction.