Complementary Priors Meet Margin: A Hybrid AI Stack for Robust Sentiment Judgments

This paper introduces an AI ensemble that brings together generative priors with strong discriminative margins, for making robust binary sentiment judgments on long-form movie reviews. Language models provide likelihood-ratio evidence while NB-SVM supplies invariant features; geometric fusion reconciles these signals. (34 words) When tested on the standard IMDB division, the hybrid stack obtains state-of-the-art accuracy for its time. Moreover, it also demonstrates measurable complementarity - weaker standalone generative paths consistently help boost the discriminative core when ensembling under tuned weights. The approach provides training recipes, feature construction, and other ablations that clarify when and why generative likelihoods add value to modern discriminative NLP. It represents a practical blueprint for building text-understanding agents in resource-poor settings.