NativeScope: Relation-Localized Retrieval over Native Topology with a Correct Anchor

Key points

• Dense retrieval usually ranks text chunks by their semantic similarity to a question.1
• 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.1
• We evaluate both methods on 200 controlled document and memory records derived from QASPER and LongMemEval under a 1,024-token budget.1
• 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.1

Sources (1)

• [1]NativeScope: Relation-Localized Retrieval over Native Topology with a Correct Anchor
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 04:21 PM

  We propose NativeScope, a scope-then-rank method for queries with a known anchor and relation.

  Dense retrieval usually ranks text chunks by their semantic similarity to a question.

Extractive summary: sentences quoted from the sources.