REBot: From RAG to CatRAG with Semantic Enrichment and Graph Routing

TL;DR

This work tackles the difficulty of providing precise academic-regulation guidance by Can Tho University students. It presents REBot, an LLM-powered chatbot built on CatRAG, a hybrid retrieval framework that unifies dense retrieval with graph-based reasoning and semantic enrichment via NER, guided by a category classifier. A CTU-specific regulation dataset and a Regulation-Enriched Knowledge Graph underpin the approach, with results showing CatRAG achieving near 98.9% F1 and real-time performance. The paper also delivers a web-based REBot application and discusses limitations and avenues for optimization, highlighting the value of integrating retrieval and graph-based reasoning in domain-specific advisory systems.

Abstract

Academic regulation advising is essential for helping students interpret and comply with institutional policies, yet building effective systems requires domain specific regulatory resources. To address this challenge, we propose REBot, an LLM enhanced advisory chatbot powered by CatRAG, a hybrid retrieval reasoning framework that integrates retrieval augmented generation with graph based reasoning. CatRAG unifies dense retrieval and graph reasoning, supported by a hierarchical, category labeled knowledge graph enriched with semantic features for domain alignment. A lightweight intent classifier routes queries to the appropriate retrieval modules, ensuring both factual accuracy and contextual depth. We construct a regulation specific dataset and evaluate REBot on classification and question answering tasks, achieving state of the art performance with an F1 score of 98.89%. Finally, we implement a web application that demonstrates the practical value of REBot in real world academic advising scenarios.

Figures (3)

• Figure 1: Knowledge graph of REBot with 250 nodes
• Figure 2: The REBot framework
• Figure 3: The Interface of our REBot System

Theorems & Definitions (1)

• definition thmcounterdefinition: REG