Title: REBot: Advancing from RAG to CatRAG through Semantic Enrichment and Graph Routing

Abstract:

Providing students with guidance on institutional policies is crucial for ensuring they understand and adhere to academic regulations. However, developing robust systems for this purpose often necessitates access to specialized regulatory resources. To overcome these hurdles, we introduce REBot, an LLM-driven advisory chatbot that leverages CatRAG, a novel hybrid retrieval-reasoning framework. This framework combines retrieval-augmented generation with graph-based reasoning capabilities. Central to CatRAG is a hierarchical, category-labeled knowledge graph, which is enriched with semantic features to ensure precise domain alignment. The system employs a lightweight intent classifier to direct queries toward the most suitable retrieval modules, thereby guaranteeing both factual precision and contextual richness. We have developed a regulation-specific dataset to evaluate REBot on classification and question-answering tasks, where it achieved state-of-the-art results with an F1 score of 98.89%. Furthermore, we deployed a web application to illustrate the practical utility of REBot in real-world academic advising contexts.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC