What Data for This Category Looks Like
Refining and chemical production document data mainly comes from unit process specifications, daily operation records, safety operation procedures, material balance reports, equipment maintenance ledgers, and similar sources. Update cycles are split into scheduled and real-time: process specifications and safety regulations are revised every six months to one year. Operation records are archived monthly. Equipment ledgers are updated in real time during equipment inspections. Most document structures follow a chapter-based format, including fields such as equipment numbers, process parameters, and operation steps. Fixed parameter units include MPa, ℃, cubic meters per hour, tons per day, and other standard units.
Constraints Imposed on Knowledge Base Retrieval and Recall
These characteristics create four constraints for retrieval and recall. First, long documents make up a large share. Preserve the integrity of process steps during chunking to avoid splitting that breaks operation logic. Second, fields include fixed professional units. Retrieval must accurately match parameter units to prevent incorrect recall from fuzzy matching. Third, data is isolated by business domain. Limit the retrieval scope to the knowledge base of the current refining and chemical project, excluding unrelated documents from other units. Fourth, the volume of historical archived data is large. Set a reasonable recall upper limit to avoid context overflow.
How to Set Configurations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Refining process parameters require precise matching to avoid low-similarity unrelated documents being included in retrieval results |
limit | Top 8–12 results | Adapts to the single-document length of refining and chemical documents, controls the total token count of the context to avoid exceeding model limits |
searchMode | Exact match | Adapts to retrieval needs that require precise matching of fields such as equipment numbers and process parameters |
usingReRank | Enabled | Re-ranks initial recall results, filters out redundant documents with mismatched parameters, and improves result relevance |
datasetSearchUsingExtensionQuery | Disabled | Terminology in refining and chemical documents is fixed, so query extension is not needed to avoid introducing ambiguity |
collectionIds | Bind the knowledge base collection ID of the current refining and chemical project | Isolates knowledge base data from different refining and chemical projects, ensuring only documents from the target business domain are retrieved |
Three Common Misconfigurations
- Symptom: After restarting the FastGPT container, the bound refining and chemical knowledge base collection ID is empty, and workflow configurations are lost. Cause: The knowledge base data directory was not mounted as a persistent volume, and local storage data was cleared when the container restarted.
- Symptom: Retrieved documents include a large number of general chemical documents unrelated to the target refining and chemical unit. Cause: The
similaritythreshold is set below 0.75, recalling low-match redundant content. - Symptom: Single-round retrieval takes more than 10 seconds, and the interface shows request timeout. Cause: The
limitis set to more than 15 entries, andusingReRankis not enabled, causing excessive resource usage in subsequent context splicing and re-ranking steps.
How to Verify Configurations Are Correct
- Access the knowledge base management page of FastGPT v4.17.0, confirm that the bound
collectionIdscorrespond to the knowledge base collection of the current refining and chemical project, with no extra or missing collections. - Initiate a retrieval targeting refining and chemical process parameters, check whether the similarity of returned documents meets business requirements, and adjust the
similaritythreshold to match precision as needed. - Test the single-round retrieval duration, and adjust the
limitandusingReRankconfigurations based on actual business response requirements. - Verify that retrieval results include exclusive fields of the target refining and chemical unit, such as equipment numbers, material flow units, and similar fields, to confirm that exact matching is active.
The values provided are common starting points and should be measured against the reader'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.