What the data for this category looks like
Data for this category primarily comes from 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 include procurement demands, procurement plans, bidding documents, submission documents, bid evaluation reports, winning bid announcements, procurement contracts, and question and reply notices. These cover the full process from procurement project initiation to performance acceptance. Data fields include procurement entity information, project numbers, subject content, evaluation criteria, and response document requirements. There are no unified fixed non-standard fields. Adaptive mapping is required based on different source systems.
What constraints these characteristics impose on knowledge base retrieval and recall
Daily updated data sources require the knowledge base to use an incremental recall mechanism. This avoids resource waste from repeated full index execution. Mixed multi-type document structures require retrieval configurations to balance matching logic for structured and unstructured content. This ensures accurate recall of content in different formats, such as bidding document clauses and winning bid announcement results. Data from multiple systems must first complete unified field mapping. Without this, field missing or matching deviation may occur during retrieval. High-frequency scenarios for handling questions and complaints require retrieval results to quickly return relevant historical documents. The recall upper limit and reranking logic need targeted adjustments to ensure key documents are displayed first.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Government procurement documents have high professionalism. A threshold that is too low will introduce irrelevant general content, while a threshold that is too high will miss professional matches. Calibrated based on actual testing |
limit | Top 10 entries | A single request for government procurement consultation involves many document types. 10 entries can cover most relevant content and avoid result overload that hinders review |
searchMode | vector+keyword | Balances semantic matching and precise keyword matching, adapting to mixed structured and unstructured document types such as bidding documents and winning bid announcements |
usingReRank | Enabled | Improves the ranking weight of highly relevant documents, ensuring key business documents such as bid evaluation reports and question reply notices are recalled first |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands exclusive terms in the government procurement field, for example expanding "winning bid" to related expressions such as "winning bid announcement" and "winning bid notification", improving matching coverage |
collectionIds | Bind corresponding datasets by business domain | Distinguishes different document collections such as procurement demands, contracts, and question reply notices, avoiding invalid interference from cross-domain documents |
Three common errors
- Phenomenon: Image paths fail to load in Markdown-format uploaded documents. Returned fields are empty or show 404 errors. Cause: External resource storage permission for the knowledge base is not configured, or images are not synchronized to the storage path bound to the knowledge base during upload.
- Phenomenon: Expected bidding document content does not appear in retrieval results. The number of returned entries is far lower than expected. Cause: The
limitparameter is set too low, orsearchModeis only configured as pure vector mode without enabling keyword matching. - Phenomenon: Only one option appears in the model selection interface. Custom text understanding models cannot be added. Cause: Multi-model access permission is not enabled in system configuration, or deployment mapping for exclusive models in the government procurement field is not completed.
How to confirm the configuration is complete
- Upload a test procurement demand document. Check that chunked content is complete, with no truncation or format errors.
- Initiate a test consultation containing "winning bid announcement". Confirm that retrieval results include the corresponding document and ranking meets expectations.
- View the knowledge base update log. Confirm that daily incremental synchronization tasks run normally with no failed records.
- Check the configuration parameter panel. Confirm that switches such as
usingReRankanddatasetSearchUsingExtensionQueryare enabled.
The values provided here 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.