What the data for this category looks like
The daily documents for this category include maintenance plans, inspection logs, defect handling forms, road construction permits, and road property damage compensation forms. Data sources are highway maintenance management systems, bridge and tunnel monitoring systems, road administration inspection systems, and traffic flow monitoring data. The data update frequency is daily. Most document structures include fields such as facility number, location, time, and handling result. Units involve road-related measurement standards such as kilometers, square meters, and cubic meters.
What constraints these characteristics impose on the knowledge base retrieval and recall link
High-frequency daily updated data requires the retrieval chain to support incremental synchronization, avoiding the time cost of full retraining. Mixed storage of multiple types of documents requires setting differentiated recall weights for different documents to match query needs of different business scenarios such as maintenance and road administration. Road-related measurement fields require precise recall, avoiding result deviations caused by fuzzy matching. Documents with high real-time requirements such as inspection logs and defect handling forms need to recall the latest entries first to ensure the timeliness of office decision-making.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the semantic similarity of professional text such as road facilities and defect descriptions, avoids irrelevant results caused by too low a threshold, and avoids missing valid information caused by too high a threshold |
limit | Top 10 entries | Adapts the number of single retrieval results for scenarios such as maintenance work order assignment and road administration law enforcement, balancing sufficient information and response speed |
searchMode | hybrid | Combines keyword and semantic retrieval, covering precise fields such as road construction permit numbers and defect locations, as well as business description text |
usingReRank | Enabled | Performs secondary sorting on initial recall results, corrects semantic matching deviations, and improves the relevance of results related to defect handling and maintenance plans |
datasetSearchUsingExtensionQuery | Enabled | Generates extended queries based on the original question, covering synonymous expressions of maintenance terminology and road property management, improving recall coverage |
collectionIds | Configured by business group | Distinguishes different knowledge base collections such as maintenance, road administration, and construction permits, accurately matching query needs of corresponding business scenarios |
Three common mistakes
- Debugging page shows recallable results, but API calls return no matching results: The cause is that the corresponding knowledge base collection ID is not correctly bound during API calls, or the passed query parameters do not include business scenario restriction fields
- Knowledge base document training fails and one-click retraining is unavailable: The cause is that the batch retraining function is not used, and single documents are repeatedly uploaded directly, or metadata of abnormal documents is missing
- Exporting knowledge base records stops after reaching 50,000 entries: The cause is that full export parameters are not configured, or the export chain is not adapted to the large volume of road-related documents in this category, and the single export limit is not adjusted
How to confirm the configuration is correct
- Call the API with a test question, compare the recall results on the debugging page, and confirm that the returned document types match the business scenario
- View the knowledge base synchronization log, confirm that daily updated documents have been successfully synchronized to the retrieval chain with no abnormal errors
- Trigger a batch retraining operation, verify whether the training status of abnormal documents is updated to successful
- Adjust the
similarityparameter range, test the number of recall results under different thresholds, and confirm that they meet business requirements
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.