HippoRAG-Edu: Hippocampus-Inspired Memory-Graph Retrieval-Augmented Generation for Multimodal Cognitive–Affective–Behavioral Trajectory Modeling and Context-Aware Intelligent Tutoring

Large language model (LLM)-based tutoring systems can generate fluent explanations, hints, and Socratic questions, but pedagogical coherence requires each utterance to be grounded in a learner’s longitudinal cognitive, affective, and behavioral history. This paper presents HippoRAG-Edu, a hippocampus-inspired memory-graph retrieval-augmented generation framework. A gated multimodal encoder represents learner episodes; an OpenIE-style indexer organizes episode, concept, and context nodes; Personalized PageRank retrieves learner-relevant episodes; and offline replay, neurogenesis, Ebbinghaus decay, and pruning maintain the graph over time.