What the data for this category looks like
Data sources for this category include tender announcements, tender documents, enterprise qualification libraries, and supplier quotations. The update frequency is once per order. Daily processed documents include tender documents, qualification certificates, business deviation tables, and technical deviation tables. Document structures follow tender format requirements and qualification response clauses, and include project numbers, qualification certification documents, quotation details, and clause deviation explanations. There are no unified fixed units, and values are adjusted based on the specific requirements of the corresponding tender project.
What constraints these characteristics impose on knowledge base retrieval and recall
Data for this category is scattered and updated per order. This requires the knowledge base to support incremental synchronization across multiple data sources, to avoid excessive time spent on full updates. Documents are structured around tender formats and qualification responses, so clauses must accurately match the numbers and content of tender requirements. This means retrieval must strengthen the ability to associate and recall clause numbers. Document structures for single tender projects are fixed but vary widely. This requires support for filtering recall results based on the current project’s format rules. Since the consequences of errors are severe, recall results must ensure high matching accuracy for key qualifications, quotations, and deviation clauses, and avoid missing core content.
How to set the configurations
These configurations are formulated based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | High confidence is required for matching tender document clauses and qualifications. Filter low-matching irrelevant content to avoid interfering with core responses. |
limit | Top 4–6 results | Cover multi-dimensional information including qualifications, quotations, and deviation clauses, while controlling context length to fit the response requirements of tender documents. |
usingReRank | Enabled | Perform secondary ranking on initially recalled results, prioritize matching clause numbers and content specified in tender requirements to improve accuracy. |
datasetSearchUsingExtensionQuery | Enabled | Expand relevant tender clauses and qualification keywords based on user questions, adapting to fuzzy query needs in tender scenarios. |
collectionIds | Specify based on the knowledge base collections associated with the current tender project | Only retrieve data sources such as qualification libraries and tender announcements involved in the current project, avoiding interference from irrelevant knowledge base content. |
Three Common Misconfigurations
- Phenomenon: No document fragments fully matching the user’s question appear in the retrieval results, and the number of returned results is 0. Cause: The knowledge base collection ID associated with the current tender project is not specified. The retrieval scope covers irrelevant historical knowledge bases, causing target content to be filtered out.
- Phenomenon: Supplementary content entered during a conversation is not saved to the knowledge base, and cannot be retrieved in subsequent queries. Cause: The automatic synchronization of conversation content to the knowledge base configuration is not enabled, or the dedicated knowledge base addition interface is not used to submit supplementary information.
- Phenomenon: When the API is called with
limitset to a value greater than 6, the returned results are still only 6. Cause: The default retrieval upper limit parameter of the privately deployed FastGPT instance has not been adjusted, so the customlimitsetting does not take effect.
How to Confirm the Configuration is Correct
- Submit a query that includes a tender clause number, check the matching accuracy of the recalled results, and adjust the corresponding configuration items to the range that meets business requirements.
- Specify the knowledge base collection ID associated with the current tender project, and check whether the retrieval results only cover documents within this collection.
- After enabling the re-ranking function, compare the recall result rankings before and after enabling to confirm that the secondary ranking logic is active.
- Call the API to set a custom number of retrieved entries, and check whether the number of returned results matches the set value.
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 2026-10-19.