A Cognitive Multimodal Framework for Automated Summary Assessment and Learning Performance Analysis

The rapid expansion of digital learning content has made educational material abundant yet cognitively overwhelming for learners. While advancements in automatic speech recognition, summarization, and assessment have emerged independently, few unified frameworks exist to optimize the learner’s cognitive experience end-to-end. Prior works often address isolated tasks, transcription, summarization, or quiz generation without modeling the pedagogical impact or personalization potential across the pipeline. In this work, the authors propose EduSynth, a multimodal cognitive framework for AI-driven summarization assessment and learning analytics. EduSynth integrates transcription, punctuation restoration, semantic chunking, abstractive summarization, blog generation, quiz synthesis, multilingual translation, and learning outcome classification, unified through a novel cognitive load estimation module. The framework is built to be modular, interpretable, and sensitive to learner effort, adapting outputs based on language, complexity, and content density. The need for such a framework stems from the growing demand for personalized, compressed, and assessment-driven learning on a scale. EduSynth directly addresses this by balancing content fidelity with cognitive relief, using metrics like entropy, readability, and syntactic depth. Empirical evaluations demonstrate high ROUGE (Recall-Oriented Understudy for Gisting Evaluation) scores, Bloom-level quiz alignment, significant reductions in estimated cognitive load, and cross-lingual fidelity with BLEU (Bilingual Evaluation Understudy) and COMET (Crosslingual Optimized Metric for Evaluation of Translation) scores. EduSynth offers a scalable foundation for intelligent educational systems.