Iterative Dual-Model Alignment for Story Evaluation

Large language models (LLMs) can both evaluate and explain text quality; however, most existing evaluators operate as static classifiers and lack the ability to refine their reasoning through interaction. We propose an Iterative Alpha-Beta Learning framework that jointly trains two complementary 8B models: an Alpha (α) classifier that assesses pairwise story engagement, and a Beta (β) generator that produces structured, rubric-guided comparative explanations. The two models co-evolve within a closed feedback loop: α provides probabilistic preference signals to guide β's Direct Preference Optimization (DPO), while β's improved explanations are reintegrated to retrain α via a KL-based contrastive objective. This dual optimization enables mutual learning: α gains interpretability and robustness from β's textual rationales, while β acquires stronger alignment and discriminative precision from α's confidence deltas. Experiments on human-annotated storypair datasets (HANNA) show that the proposed system consistently outperforms strong singlemodel baselines in both accuracy and explanation quality across multiple iterative rounds 1 .