What the data for this category looks like
This category of data originates from MES systems, online inspection equipment, laboratory test results, process parameter libraries, customer quality standards, and historical defect libraries. Data updates are collected continuously alongside production batches and work shifts, with no unified fixed cycle. Daily processed documents include control plans, patrol inspection records, SPC reports, finished product inspection reports, nonconforming product review forms, release forms, and quality monthly reports. Fields include inspection batch identifiers, nonconforming item codes, release thresholds, process parameter ranges, and more. Some documents include quantified test values and associated units.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The data characteristics of this category create multiple constraints for the retrieval and recall process. Multi-source, frequently updated data requires the system to support incremental data recall to match real-time synchronization needs for each batch and shift. Diverse document types and varied field structures require retrieval logic to adapt to the semantic traits of different documents, preventing loss of key inspection item information due to format differences. Quantified thresholds and test values demand precise matching of fields and units during retrieval, to avoid nonconforming product release or misjudgment caused by semantic ambiguity. The associated retrieval requirements of the historical defect library mean recall results must enable cross-document linking to processing records for similar issues.
How to configure the setup
This document is compatible with FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets semantic precision requirements for matching inspection standards and defect records; avoids overly low values that return irrelevant results, and overly high values that miss valid references |
limit | Top 8 entries | Balances retrieval result comprehensiveness and context window usage, adapting to the multi-document reference needs of quality audits |
searchMode | mixedRecall | Covers both quantified value matching and semantic association retrieval, supporting multiple document types such as SPC reports and control plans |
usingReRank | Enabled | Performs secondary sorting on initial recall results, corrects semantic matching deviations, and improves the accuracy of nonconforming item associated retrieval |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval terms based on the original query, covering precise matching of fields including process parameters and batch numbers |
collectionIds | Bind exclusive collections by business domain | Isolates production quality data from other business data, preventing irrelevant documents from interfering with retrieval results |
Three common configuration errors
- Retrieval results returned by the API do not match those from the platform test interface. The cause is failure to specify the
collectionIdsparameter in the API request. The default configuration loads the full knowledge base content, and does not bind the exclusive data collection for the current business domain. - Unable to quickly adjust retrieval index parameters through the retrieval test interface. The cause is failure to access the content management page of the corresponding knowledge base, and directly modifying the index via the edit entry in the retrieval test page. This entry only supports basic configuration adjustments.
- Retrieval results include downloadable original document attachments. The cause is failure to disable document download permissions in the knowledge base configuration, which allows users to access original quality documents through retrieval results.
How to verify the configuration is complete
- Confirm that API request parameters match the
collectionIdsconfigured on the platform, ensuring the called knowledge base scope matches that of the test interface. - Submit a test query containing inspection batch numbers and nonconforming item codes, to verify that retrieval results include document content from the corresponding business domain.
- Check that the number of retrieval results aligns with the
limitparameter configuration, with no excess or insufficient entries. - Confirm that no download links for original documents appear in retrieval results, verifying that the document permission configuration is active.
The values provided are common starting points and should be measured against the reader's own samples.