PolyUQuest: Verifiable Structure-Aware Web RAG over Heterogeneous Graphs

2026-07-11 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, long

Summary

PolyUQuest is a novel, verifiable, structure-aware web Retrieval-Augmented Generation (RAG) framework designed to overcome limitations of existing systems that treat web content as flat text. Developed by researchers at The Hong Kong Polytechnic University, it constructs a heterogeneous graph unifying hyperlink topology, DOM hierarchy, and entity-relation knowledge across web pages. A two-tier router intelligently dispatches queries to one of three retrieval modes: direct block retrieval, cross-page graph traversal, or multi-hop entity reasoning. This system ensures full answer verifiability, providing source page, heading path, and entity links for every cited block. Evaluated on 4,240 PolyU web pages, PolyUQuest achieved superior correctness (0.644), coverage (0.649), and faithfulness (0.921), with a 36-point faithfulness gain over the next-best baseline. It also significantly reduced LLM token consumption to 2,968 tokens per query, demonstrating efficiency. The system is slated for deployment as a student-facing QA service.

Key takeaway

For Machine Learning Engineers developing RAG systems over complex web content, PolyUQuest demonstrates that integrating hyperlink topology, DOM hierarchy, and entity-relation knowledge into a unified heterogeneous graph is crucial. Your team should implement a query routing mechanism to dispatch requests to structure-matched retrieval modes, significantly improving answer correctness, coverage, and faithfulness while reducing LLM token costs. Prioritize verifiable provenance by linking answers directly to source blocks and paths, enhancing user trust and debugging capabilities for institutional or enterprise deployments.

Key insights

PolyUQuest unifies web structure into a heterogeneous graph for verifiable, efficient, and accurate RAG.

Principles

• Unify web structure: hyperlinks, DOM, entities.
• Route queries to structure-matched retrieval modes.
• Provide full provenance for answer verifiability.

Method

PolyUQuest's offline phase builds a three-layer heterogeneous graph from crawled web pages. The online phase routes queries via a two-tier router to structure-aware retrieval modes (direct block, navigation, entity reasoning) for evidence assembly and answer generation.

In practice

• Implement structure-aware RAG for complex websites.
• Enhance QA systems with verifiable answer provenance.
• Optimize LLM token usage via query routing.

Topics

• Retrieval-Augmented Generation
• Heterogeneous Graphs
• Web Structure Analysis
• Query Routing
• Verifiable AI
• Question Answering Systems

Code references

• 13-pieces-teen/PolyUQuest
• circlemind-ai/fast-graphrag

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• PolyUQuest: Verifiable Structure-Aware Web RAG over Heterogeneous Graphs — PolyUQuest enhances RAG by using heterogeneous graphs for structure-aware, verifiable web information retrieval.
• Architectural patterns for graph-enhanced RAG: Moving beyond vector search in production — Graph-enhanced RAG improves LLM accuracy by integrating structural data from graph databases with semantic vector search.
• FAIR GraphRAG: A Retrieval-Augmented Generation Approach for Semantic Data Analysis — Integrating FAIR Digital Objects into graph-based RAG significantly enhances data findability, interoperability, and question-answering performance.
• Don’t Put Everything into RAG: Building a Safe Candidate Layer for Web Sources — Directly feeding unverified internet data into RAG systems introduces significant risks and should be avoided.
• Proxy-Pointer RAG: Solving Entity and Relationship Sprawl in Large Knowledge Graphs — Proxy-Pointer RAG uses structural document context for accurate entity and relationship reconciliation in knowledge graphs.
• Embeddings Aren’t Magic: The Predictable Failure Modes of RAG Retrieval — Embeddings provide synonym-tolerant search but predictably fail on structural issues like OOV terms and logical operations.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.