Back to AI Research

AI Research

PolyUQuest: Verifiable Structure-Aware Web RAG over... | AI Research

Key Takeaways

  • PolyUQuest is a research framework designed to improve how AI systems retrieve and generate information from complex websites.
  • Existing retrieval-augmented generation (RAG) systems treat web pages as flat text, losing the structural and semantic signals encoded in HTML.
  • We present PolyUQuest, a verifiable, structure-aware web RAG framework built on a heterogeneous graph that unifies hyperlink topology between pages, DOM hierarchy within pages, and entity-relation knowledge across pages.
  • A two-tier router dispatches each query to one of three retrieval modes matched to its structural need, including direct block retrieval, cross-page graph traversal, and multi-hop entity reasoning.
  • Every answer is fully verifiable, as each cited block carries its source page, heading path, and entity links so that users can trace any claim back to its structural evidence.
Paper AbstractExpand

Existing retrieval-augmented generation (RAG) systems treat web pages as flat text, losing the structural and semantic signals encoded in HTML. We present PolyUQuest, a verifiable, structure-aware web RAG framework built on a heterogeneous graph that unifies hyperlink topology between pages, DOM hierarchy within pages, and entity-relation knowledge across pages. A two-tier router dispatches each query to one of three retrieval modes matched to its structural need, including direct block retrieval, cross-page graph traversal, and multi-hop entity reasoning. Every answer is fully verifiable, as each cited block carries its source page, heading path, and entity links so that users can trace any claim back to its structural evidence. We evaluate on the official websites of the Hong Kong Polytechnic University (PolyU), comprising 4,240 pages, 31,086 DOM blocks, 29,119 entities, and 37,680 relations, together with a multi-type evaluation benchmark. PolyUQuest outperforms existing RAG systems in answer correctness, coverage, and faithfulness, while consuming significantly fewer LLM tokens per query. The demonstration provides an interactive interface for inspecting cited answers, comparing retrieval traces across routing modes, and exploring evidence graph paths. PolyUQuest is being prepared for deployment as a student-facing QA service at PolyU.

PolyUQuest is a research framework designed to improve how AI systems retrieve and generate information from complex websites. Traditional Retrieval-Augmented Generation (RAG) systems often treat web pages as simple, flat text, which causes them to lose critical structural information like how pages are linked or how content is organized within a document. PolyUQuest addresses this by using a "structure-aware" approach that treats a website as a unified, heterogeneous graph, allowing the AI to navigate between pages, understand document hierarchies, and connect related entities to provide more accurate and verifiable answers.

A Three-Layer Web Graph

To better understand a website, PolyUQuest organizes information into three distinct layers. First, it maps the "site graph," which tracks hyperlinks between different pages. Second, it builds a "block tree" for each page, which preserves the original document hierarchy—such as headings and sections—instead of breaking text into arbitrary chunks. Finally, it creates an "entity graph" that links specific people, programs, departments, and research topics across the entire site. By combining these layers, the system can trace a path from a specific entity to the exact block of text where it appears, and then to the broader page context.

Intelligent Query Routing

Not every question requires the same search strategy. PolyUQuest uses a two-tier router to analyze a user's query and dispatch it to one of three specialized retrieval modes:

  • Direct Block Retrieval: Used for simple, single-page factual questions.

  • Navigation Retrieval: Used for complex queries that require aggregating information across multiple linked pages.

  • Entity Reasoning: Used for multi-hop questions that require connecting different concepts or people across various parts of the website.
    This targeted approach ensures the system only retrieves the most relevant information, which significantly reduces the number of tokens (computational units) required by the AI compared to systems that process large amounts of irrelevant global context.

Verifiable Answers and Provenance

A key feature of PolyUQuest is its focus on transparency. Every answer generated by the system includes citations that link directly back to the source. Users can click these citations to see the specific DOM block, the heading path, and the source page that provided the evidence. This allows users to verify the AI's claims instantly. The system also provides an interactive interface where users can visualize the "evidence graph," helping them see the path the AI took to find the answer and explore related connections within the website.

Performance and Results

When tested on the official website of The Hong Kong Polytechnic University (PolyU), which includes over 4,000 pages and tens of thousands of entities, PolyUQuest outperformed existing RAG systems in correctness, coverage, and faithfulness. Notably, it achieved a significantly higher faithfulness score—meaning its answers were more reliably supported by the retrieved evidence—while using substantially fewer tokens per query than other graph-based RAG methods. The research demonstrates that preserving the structural hierarchy of web pages is the most important factor in improving the quality of AI-generated answers for complex, site-specific queries.

Comments (0)

No comments yet

Be the first to share your thoughts!