HAPI: A Hybrid Knowledge Graph and Vector Search Framework for Efficient Vietnamese Traffic Law Retrieval

Abstract

The rapid increase in personal vehicles in Vietnam has made traffic law enforcement a critical issue. However, accessing accurate legal information remains a challenge for citizens due to the complexity of legal terminology. Traditional keyword-based search engines often fail to understand natural language queries, while modern Large Language Models (LLMs) suffer from high latency and high operational costs. To address these limitations, this paper proposes a hybrid retrieval framework for Vietnamese traffic law. Our approach combines the deterministic accuracy of Knowledge Graphs (KG) with the semantic understanding of Vector Search, further optimized by a Cross-Encoder reranker. Experimental results on a newly constructed dataset (VTL-200) demonstrate that our system achieves an Accuracy@1 of 0.565 and an outstanding Accuracy@5 of 0.93. Furthermore, the proposed method significantly reduces inference latency (averaging 4.06 seconds) and computational costs compared to state-of-the-art LLM-based Retrieval-Augmented Generation (RAG) baselines such as Qwen-3 and Gemini 2.5-Flash. The code and dataset will be open-sourced upon acceptance.