Context-Aware Sentiment Validation Framework for Cinematic Emotion Understanding Using Hierarchical Narrative Modeling and Causal Attributions

Abstract

The need to develop film analysis models capable of analyzing complex sentiment in cinematic stories requires moving past binary sentiment classifications towards more cognizant, emotionally relevant storytelling models. Current approaches have relied upon simplistic, unidirectional sequence-based models or static, non-hierarchical embedding representations. These limitations lead to diminished interpretability and inconsistent prediction performance in IMDb review based applications. As an alternative, we are proposing a Sentiment Validation Framework that integrates five new modules: Contextual Narrative Embedding with Hierarchical Scene Fusion (CNE-HSF), Emotion Transition Graph Attention Network (ET-GAT), Causal Sentiment Attribution via Counterfactual Narrative Masking (CSA-CNM), Multi-Dimensional Emotion Calibration with Latent Polarity Alignment (MEC-LPA), and Trust-Aware Sentiment Validation with Narrative Consistency Index (TSV-NCI). Together these modules enhance the fidelity of contextual representation, the ability to reason about causality, and provide an index of trust regarding the narrative consistency of the predicted sentiment, allowing for a comprehensive, robust, and explainable, narrative consistent sentiment model. Therefore, the proposed framework can be used as a basis for developing advanced cinematic analytics tools and intelligent content evaluation systems.