Extracting Agricultural Experts' Experiential Knowledge from Non-Standardized Text with Quantity and Operation Scope Constraints

Expert advice on crop management is often written informally, and in such text a number can easily be assigned the wrong role or attached to the wrong operation, material, or measurement basis. We propose QS-IE, a knowledge extraction method constrained by quantity and operation scope. Its unit of extraction is a complete quantitative fact. Given a shared set of numeric candidates and a single base model, QS-IE selects complete candidates from several prompting strategies, verifies them against the source text, and applies scope constraints, so that every extracted quantity stays linked to its source evidence. We built a corpus of 4006 texts on tomato cultivation. Two researchers with agronomy backgrounds jointly annotated and reviewed 1000 of them; the other 3006 were annotated with large language model assistance and then reviewed by hand. The corpus was split into 3206 development texts and 800 independent test texts. On the 1152 reference facts of the test set, QS-IE reached a precision of 84.02%, a recall of 86.72%, and a micro-averaged F1 of 85.35%, which is 3.91 percentage points higher than the strongest baseline, a multi-strategy in-context learning ensemble. Removing the operation scope constraints lowered test F1 by 8.24 percentage points. In a sampled human audit of the test set, 82.00% of the outputs for texts containing target facts were judged usable, and all six engineering checks on storage and evidence traceability were passed. The results indicate that explicit association constraints make quantitative facts extracted from tomato cultivation texts more complete and easier to review, which supports the structured, traceable storage of experiential knowledge that experts express in non-standardized text.