What the data for this category looks like
Data for this category comes from national public resource trading platforms, local public resource trading platforms, engineering approval systems, and bidding and tendering supervision systems. It updates once per day. Daily documents include bidding announcements, qualification pre-qualification documents, bidding documents, tender documents, bid evaluation reports, public notices of successful bidders, successful bid result announcements, and more. Fields covered include section division, qualification requirements, bills of quantities, bid evaluation expert records, objection and complaint handling records, and others. Some documents contain structured tables and image content.
Constraints on Knowledge Base Retrieval and Recall
Data for this category has a high update frequency, multiple document types, and large structural differences. This requires the retrieval link to support incremental synchronization and adaptation to multiple document types. Core retrieval needs vary significantly across different documents: bidding announcements require precise matching of section numbers, bid evaluation reports require recalling expert records and bid evaluation rules, and objection and complaint handling requires associating historical successful bid notices and qualification requirements.
Unlike government procurement scenarios, documents in this category often involve structured content such as qualifications, performance records, and bills of quantities. Retrieval must balance field matching and text understanding. Business errors directly relate to retrieval accuracy. Irrelevant documents must be excluded from retrieval results. Recalled content must be strongly relevant to the current project.
How to Set Configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Engineering bidding documents require a high matching threshold to avoid irrelevant historical announcements from appearing in current project retrieval results |
limit | Top 10 | Internal office scenarios such as bid evaluation and objection handling need a limited number of precise results to prevent excessive information from interfering with decision-making |
searchMode | Vector Retrieval + Keyword Hybrid | Bidding documents have both structured fields (such as section numbers and bills of quantities) and natural text content. The hybrid mode balances recall rate and accuracy |
usingReRank | Enabled | Long-text documents such as bid evaluation reports and tender documents need secondary ranking to improve the relevance of core business content |
datasetSearchUsingExtensionQuery | Enabled | User questions such as objection consultations and qualification queries can be expanded into associated retrieval terms to cover more relevant documents |
collectionIds | Bind corresponding collections by business scenario | Manage bidding documents, bid evaluation documents, and objection documents in separate collections to avoid cross-category data interfering with retrieval results |
Three Common Mistakes
- Phenomenon: An error occurs when calling knowledge base retrieval, and the log shows document parsing failure. Cause: No dedicated parsing rules are configured for multiple document types such as bidding announcements and bid evaluation reports. The mixed format causes the parsing module to fail to process correctly.
- Phenomenon: Irrelevant historical successful bid announcements account for an overly high proportion of retrieval results, and it is impossible to focus on the qualification requirements of the current section. Cause: The
similaritythreshold is not set or is set too low, andusingReRankis not enabled for secondary filtering, resulting in low-relevance content being recalled first. - Phenomenon: Images in bid evaluation reports in the local knowledge base cannot be retrieved, and only plain text content is returned. Cause: The document image OCR parsing plugin is not enabled. Only plain text data in the document is extracted, and key information such as bid evaluation records and expert signatures in the images is not processed.
How to Confirm Configurations Are Correct
- Manually upload the bidding documents and corresponding bid evaluation reports for the current section, trigger a retrieval for the section's qualification requirements, and check whether the returned results include the target documents and the relevance meets expectations.
- Check the knowledge base synchronization log to confirm that the number of daily incrementally updated documents meets business expectations, with no abnormal errors or unsynchronized documents.
- Test retrieving a bid evaluation report containing images, and confirm that the returned results include text content extracted via image OCR, covering key information in the images.
- Adjust the
similaritythreshold, observe changes in the number of retrieval results, and confirm that the threshold setting matches the accuracy requirements of the current business scenario.
The values listed here are common starting points and should be evaluated 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.