What Data for This Category Looks Like
Document data for oilfield engineering services mainly comes from completion reports, technical disclosure documents, compliance review documents, and operation guides for operations including drilling, fracturing, cementing, and oil testing. Data updates follow individual oilfield project cycles. Bulk archives are created after project completion, with partial adjustments made daily as compliance clauses are revised and operation standards are updated. Most documents use structured formats, with fixed fields including well number, operation type, construction parameters, approval nodes, and compliance clauses. Well numbers and operation numbers act as auxiliary metadata. Units include cubic meters, megapascals, hours, and other engineering-specific units.
Constraints for Knowledge Base Retrieval and Recall
Structured fields in oilfield engineering documents require retrieval to precisely match unique identifiers such as well numbers and operation numbers, to avoid mixing documents across projects. Documents contain a large number of technical terms and specific engineering parameters. Recall must prioritize retaining semantically relevant professional content, to prevent generalized expressions from interfering with document drafting accuracy. The project-based archiving feature requires strict limitation of retrieval scope to the document set of the current project, to prevent irrelevant data from interfering with results. Most documents are long texts, so retrieval chunk length must adapt to complete expression of engineering parameters, to avoid truncating key construction information.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the technical term density of oilfield engineering documents, balances recall coverage and relevance filtering |
limit | Top 8 entries | Covers typical citation needs for a single engineering document, avoids overly long contexts exceeding model processing limits |
searchMode | Precise matching | Targets structured fields such as well numbers and operation numbers, prioritizes recalling documents with exact matches |
usingReRank | Enabled | Re-ranks initial recall results by professional relevance, corrects results that are semantically similar but do not match the operational scenario |
datasetSearchUsingExtensionQuery | Enabled | Automatically supplements extended queries for operation types and well block ranges, improves coverage of cross-field retrieval |
collectionIds | Configured per project-specific collection | Limits retrieval scope to the document set of the current oilfield project, aligns with the collection permission logic of FastGPT v4.17.0 |
Three Common Misconfigurations
- Issue: Returned answer content does not match the documents cited in the knowledge base, and citation fields point to documents unrelated to the answer. Cause:
usingReRankis not enabled, initially recalled low-relevance documents are incorrectly used as the basis for the answer, and citation associations are not verified. - Issue: The number of documents returned by retrieval is far lower than expected. Cause: The
similaritythreshold is set too high, filtering out some engineering documents that are semantically relevant but have slightly different term expressions. - Issue: Retrieval results include irrelevant documents from other projects. Cause:
collectionIdsis not configured correctly, and the retrieval scope covers knowledge base sets that do not belong to the current project.
How to Verify Proper Configuration
- Upload a test oilfield cementing operation document, run a query containing "cementing" and a test well number. Check whether the test document is included in the recall results.
- View the citation fields of retrieval results, confirm that each answer corresponds to a specific document in the knowledge base, with no empty citations or irrelevant citations.
- Adjust the
similaritythreshold to 0.8, compare the number of recalled documents before and after the adjustment, confirm that the threshold adjustment produces the expected relevance filtering effect. - Enable
datasetSearchUsingExtensionQuery, compare the number of recall results before and after enabling, confirm that the extended query takes effect.
The values listed here are common starting points, and should be evaluated against the user's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.