Native unit retrieval improves question answering over structured documents
NativeScope: Relation-Localized Retrieval over Native Topology with a Correct Anchor
Information RetrievalComputation and Language
Summary
Finding helpful information in long documents is hard because most systems only look at how similar text parts are to a question, ignoring how the document is organized. The authors propose a new way called NativeScope that uses known points in the document and their structural relations to narrow down where to look before ranking text parts by relevance. This approach works much better at finding the right parts in documents and memory records when the position and relations are known accurately. However, if the starting point is wrong, the method’s accuracy drops significantly.
What this means in practice
- •For enterprise search teams: Improve retrieval accuracy by using document structure and known reference points to narrow search scopes before ranking results.
- •For knowledge base developers: Enable more precise answer finding in large, structured memory or document collections by leveraging anchors and relations for scoped retrieval.
Authors
Long Wang
Abstract
Dense retrieval usually ranks text chunks by their semantic similarity to a question. This ignores structure that many data systems already store, including section membership, session boundaries, and native order. We propose NativeScope, a scope-then-rank method for queries with a known anchor and relation. It represents a query as q -> (A, r, B). The anchor A and relation r select native units through belonging, before, or after operators, and the target term B ranks only chunks that overlap the selected scope. An internal variant, NS-FullQ, ranks the same candidates with the full question. We evaluate both methods on 200 controlled document and memory records derived from QASPER and LongMemEval under a 1,024-token budget. NativeScope attains native-unit recall of 89.28 percent for documents and 72.50 percent for memories, improving over instance-wide Dense RAG by 42.75 and 22.00 percentage points. NS-FullQ reaches 87.78 percent and 68.50 percent; its differences from NativeScope are inconclusive, locating the primary gain in relational scoping rather than the shorter ranking query. With automatic Top-1 anchors, memory recall falls to 35.50 percent. NativeScope is therefore effective when anchor coordinates and native relations are reliable, but hard scoping inherits errors from the localization interface.