What the data for this category looks like
Data sources for this category include enterprise procurement platforms, public resource trading platforms, supplier databases, contract management systems, and audit issue databases. Update frequency is daily. Daily documents cover procurement plans, requirement specifications, bidding documents, bid evaluation records, contracts, acceptance documents, and supplier complaint materials. Core fields include procurement methods, budget basis, scoring methods, supplier qualification requirements, performance acceptance clauses, and more. Most documents are a mix of structured and unstructured formats, with no unified fixed structure.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Multi-source and heterogeneous data sources require precise limitation of the retrieval scope to avoid mixing irrelevant documents outside tendering and procurement compliance. The daily update feature requires the retrieval pipeline to adapt to incrementally synchronized datasets, ensuring recalled content uses the latest compliance provisions. Documents mix structured tables and unstructured text, so the system must adapt to chunking and encoding rules for different content types. Core fields involve key judgment items for procurement compliance. Retrieval must prioritize semantic matching of these fields, otherwise risks such as bid rigging and audit rectification may arise. Additionally, the party building learning Q&A scenario requires retrieval results to link procurement compliance requirements related to party building, so the semantic coverage of queries must be expanded.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Tendering and procurement compliance requires accurate matching of compliance provisions. This threshold filters low-relevance results and avoids interference with audit and procurement decisions |
limit | Top 8 entries | Balances context length and recall coverage. Prevents excessive redundant content from occupying token quotas, while covering key compliance basis |
searchMode | vector | Most tendering and procurement documents are clearly worded compliance rules. Vector retrieval can accurately match similar procurement compliance provisions |
usingReRank | Enabled | For scenarios with many similar questions, re-ranking filters recall results with semantic deviations, improving the accuracy of compliance judgments |
datasetSearchUsingExtensionQuery | Enabled | Adapts to the party building learning Q&A scenario. Extended queries can supplement party building-related procurement compliance guidelines, aligning with business needs |
collectionIds | Specify the exclusive dataset ID for tendering and procurement compliance | Limits the retrieval scope to only documents of this category, avoiding mixing irrelevant content from other categories |
Three common mistakes
- Phenomenon: Retrieval returns empty results, or the content extracted from knowledge base references is empty. Cause:
collectionIdsis not configured to limit the retrieval scope, so the retrieval covers datasets outside tendering and procurement compliance, and no matching content can be found. - Phenomenon: Similar non-compliance questions (such as system password recovery) are recalled first. Cause: The
similaritythreshold is set too low, andusingReRankis not enabled, leading to low-relevance non-compliance content interfering with retrieval results. - Phenomenon: Image content within documents cannot be output. Cause: The image extraction function for document parsing is not enabled, or the knowledge base configuration does not enable image indexing, so image content is not included in the retrieval and recall scope.
How to confirm the configuration is complete
- Manually enter compliance questions related to procurement methods and budget basis, check whether the recalled results only include content from tendering and procurement compliance documents.
- Adjust the
similaritythreshold, observe changes in the relevance of recalled results, and confirm that the accuracy requirements of the business scenario are met. - Upload test tendering and procurement documents, verify that corresponding clauses can be correctly recalled, and no irrelevant content is mixed in.
- Compare recall results with and without
usingReRankenabled, confirm that the re-ranking function works properly.
The values given 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.