What Data Does This Category Include
The data sources for this category include the China Government Procurement Network, local government procurement networks, fiscal integration systems, procurer budget systems, and supplier credit records. The update frequency is daily. Commonly involved documents cover procurement requirements, procurement plans, bidding documents, tender documents, bid evaluation reports, winning bid announcements, procurement contracts, response to inquiries, and more. Service subjects include procurer units, government procurement supervision offices of fiscal departments, procurement agencies, and evaluation experts. Documents must strictly comply with relevant government procurement regulatory rules, and include core fields such as qualification review, scoring criteria, and contract terms.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
Daily updated data requires the retrieval system to support flexible switching between incremental synchronization and full updates, to avoid audit errors caused by data lag. Multi-source and heterogeneous document formats, including text documents, spreadsheet files, and Feishu documents with images, require the retrieval system to support multiple parsing formats. Documents must strictly match regulatory rules, meaning recall results must have high confidence, and must not miss key clauses or mix in irrelevant content. Additionally, documents involve sensitive content such as supplier qualifications and scoring methods, so the retrieval scope must be strictly limited to exclusive government procurement datasets to prevent interference from cross-category data.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.88 | Core content of government procurement documents requires high matching accuracy, filter low-relevance general regulatory documents to ensure recalled content aligns with review scenarios |
limit | Top 8–12 results | Declaration materials and file review require multiple related documents for verification, such as cross-checking tender documents and supplier credit records. 8–12 results can cover core related content |
searchMode | Hybrid search | Balances semantic matching and precise keyword matching, covers government procurement-specific terminology and official standardized expressions, avoids missing key clauses |
usingReRank | Enabled | Government procurement documents are lengthy and complex in content. The re-ranking function can further filter low-relevance recall results and improve review efficiency |
datasetSearchUsingExtensionQuery | Enabled | Government procurement field has specific terms and synonyms. Extended queries can cover official standardized expressions and improve the comprehensiveness of retrieval and recall |
collectionIds | Bind exclusive government procurement dataset, single collection data size no more than 1000 MB | Strictly limit retrieval scope to prevent mixing non-government procurement documents, ensure review content complies with regulatory rules |
Three Common Errors
- Issue: Low retrieval result accuracy after importing docx and excel files, with a large number of irrelevant content returned. Cause: The
similaritythreshold was not adjusted for government procurement documents, andusingReRankwas not enabled to filter redundant results, leading to low-matching content being included in the recall set. - Issue: Images in Feishu documents stored in the knowledge base cannot be displayed normally in retrieval results. Cause: The external link access permission for Feishu documents was not configured, or the image extraction and parsing function for document content was not enabled, preventing normal calling of image links.
- Issue: A configuration error occurs when creating a new knowledge base after privately deploying FastGPT. Cause: The dataset permission corresponding to
collectionIdswas not correctly bound, or the storage path configuration during deployment did not meet system requirements.
How to Verify Proper Configuration
- Retrieve core clauses in procurement requirement documents, verify whether the returned results include related content of corresponding procurement plans and bidding documents, to confirm that the retrieval scope covers the correct dataset.
- Test importing docx, excel, and Feishu documents, confirm that all formats can be parsed normally, valid content can be recalled, and image links can load properly.
- Check the
similarityscore of retrieval results, confirm that the score falls within the preset matching range to meet the requirements of the review scenario. - After enabling
usingReRank, verify whether the sorting of returned results aligns with business logic, prioritizing highly relevant core documents.
The configuration values listed above are common starting points. They should be evaluated against the reader's own specific 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.