Researchers from Hong Kong Polytechnic University have developed PolyUQuest, a web retrieval-augmented generation (RAG) system that preserves the structural and semantic signals encoded in HTML. Unlike existing RAG systems, PolyUQuest treats web pages as heterogeneous graphs that unify hyperlink topology, DOM hierarchy, and entity-relation knowledge. This framework includes three retrieval modes matched to a query's structural need: direct block retrieval, cross-page graph traversal, and multi-hop entity reasoning. Each answer is accompanied by traceable provenance, allowing users to inspect the evidence behind claims. --- Why it matters: PolyUQuest matters because it addresses a significant limitation of existing RAG systems, which treat web pages as flat text. This can lead to loss of structural and semantic signals that are crucial for accurate and informative answers. PolyUQuest's ability to preserve these signals can improve the correctness, coverage, and faithfulness of generated answers. Source: https://arxiv.org/abs/2607.08269