What data looks like for this category
Data for this category comes from engineering orders, sand, cement, admixture inspections, moisture content tests, mixing plant control systems, and laboratories. Daily documents include mix proportion notices, batch opening inspection records, production and delivery slips, raw material inspection records, test block records, and strength reports. Update frequency follows per vehicle production, per batch of raw material delivery, and daily summaries. Data fields include mix proportion parameters, slump indicators, strength values, raw material inspection results, and similar items, using units such as millimeters and megapascals.
Constraints imposed by these characteristics on knowledge base retrieval and recall
The high-frequency updates, multiple document types, and strongly business-bound field characteristics of this category create multiple constraints for retrieval and recall. High-frequency updates per vehicle and per batch of raw materials require retrieval systems to adapt to fast synchronization logic for incremental data, avoiding recall of outdated mix proportion or inspection data. Diverse document types include multi-row table XLSX raw material records, requiring adapted chunking and recall rules for structured data. Business processes rely on precise matching of parameters with units such as slump and strength values, requiring retrieval to retain unit associations for fields to prevent parameter mismatches. Additionally, production cycles are tight, so retrieval response times must meet on-site scheduling time requirements.
How to set configurations
This configuration is based on the knowledge base feature of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85, calibrated via actual testing | Parameters for this category require precise matching. A threshold that is too low will introduce irrelevant raw material inspection records, while a threshold that is too high may miss eligible mix proportion data |
limit | Top 8–12 entries | Production scheduling and inspection audits need to reference multiple types of data such as mix proportions, raw material inspections, and test block records. 8–12 entries cover core reference information while avoiding overly long context |
searchMode | mixedRecall | This category includes structured tables and plain text documents. The hybrid retrieval mode can match both semantic relevance and precise field keywords, adapting to diverse document types |
usingReRank | Enabled | Multi-source data with high-frequency updates often produce redundant results with similar semantics. Reranking improves the relevance of the top 3 results, meeting the fast decision-making needs of production audits |
datasetSearchUsingExtensionQuery | Enabled | Business personnel queries often omit units or use abbreviated parameters. Query optimization can supplement matching standard terms such as slump and strength, improving recall accuracy |
collectionIds | Configured grouped by document type | Separating mix proportion notices, raw material inspection records, and test block reports into independent collections can precisely limit the retrieval scope, avoiding invalid cross-category recall |
Three common mistakes
- Symptom: Knowledge base retrieval takes longer than 4 seconds, with a timeout status displayed on the page. Cause: The retrieval scope was not optimized for high-frequency updated multi-source data, and a large number of redundant historical documents were recalled across collections, leading to excessive retrieval computation load.
- Symptom: Imported multi-row XLSX raw material inspection tables only return scattered text, with no structured field matching. Cause: Correct chunking rules for multi-row tables were not configured, and direct paragraph-based chunking caused structured data to be split, failing to retain field associations.
- Symptom: Retrieval results include strength values or slump data without units, which cannot meet business audit requirements. Cause:
datasetSearchUsingExtensionQuerywas not enabled, or field unit association rules were not configured, leading to recalled parameters missing necessary unit information for business use.
How to confirm the configuration is correct
- Import a single mix proportion notice and raw material inspection record, run a business query containing "adjust water dosage", and verify whether the recall results cover the target documents.
- View the retrieval log to confirm that the response time meets on-site production time requirements, with no persistent timeouts.
- Check the field completeness of retrieval results, confirming that returned content includes required business parameters and corresponding unit information.
- Import a new batch of raw material inspection data, run a corresponding query, and confirm that the new data can be recalled normally.
The values provided here are common starting points and should be measured against the reader's own samples.