What this category's data looks like
Data for this category comes from state-owned enterprise and public sector procurement platforms, public resource trading platforms, supplier databases, contract management systems, and audit issue repositories. The update frequency is daily. Daily documents include procurement plans, requirement specifications, bidding documents, bid evaluation records, contracts, acceptance documents, and supplier complaint materials. Each document type has dedicated fields. For example, procurement plans include budget basis and procurement methods. Bidding documents include scoring rule details. Bid evaluation records include expert scores and compliance judgment items. The overall data structure is layered according to procurement process nodes.
Constraints for knowledge base retrieval and recall
Daily updated data sources require the retrieval pipeline to support incremental synchronization and full refresh. This prevents data lag from affecting business processing. Mixed retrieval across multiple document types requires adapting to field weights for different documents. For example, compliance judgment items in bid evaluation records must be prioritized for recall. Data formats vary widely across different sources. A unified preprocessing workflow is required to ensure retrieval consistency. The layered document structure based on procurement process nodes requires retrieval to accurately match business links. For example, the bidding link prioritizes recalling bidding documents and procurement plans. The acceptance link prioritizes matching contracts and acceptance documents.
Configuration setup
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the high density of technical terminology in tendering and procurement documents, filters low-relevant compliance-related content |
limit | Top 6–8 results | Covers reference documents across multiple stages of the procurement process, avoids insufficient information in single results |
searchMode | Hybrid retrieval | Matches both full-text keywords and structured fields, adapts to retrieval needs across multiple document types |
usingReRank | Enabled | Reorders initial recall results by relevance, improves ranking accuracy for high-weight documents such as bid evaluation records and bidding documents |
datasetSearchUsingExtensionQuery | Enabled | Expands compliance-related retrieval terms based on business keywords entered by users, covers hidden associations across documents such as procurement plans and contracts |
collectionIds | Bind corresponding knowledge base collections by business node | For example, bind bidding documents and procurement plan collections for the bidding link, bind contract and acceptance document collections for the acceptance link, accurately narrow the retrieval scope |
Three common errors
- A
common:core.chaterror occurs during retrieval. This is caused by incorrectly configured knowledge base collections bound tocollectionIds, or documents in the collection not having completed full indexing. - Documents in multi-level directories cannot be recalled. This is caused by not enabling the directory recursive parsing switch, so documents at all levels are not included in the retrieval index range.
- The number of retrieval results does not match the set
limit. This is caused by the initial retrieved valid results not reaching the set upper limit, or the re-ranking stage filtering some low-relevant results that do not meet compliance requirements.
How to confirm correct configuration
- Manually enter keywords related to procurement plans. Check if the recalled results include documents related to procurement plans and procurement methods.
- Switch business node tags. Verify if the knowledge base collections bound to
collectionIdsautomatically switch, and if the retrieval results match the document types of the current node. - After enabling
usingReRank, compare the initial recall results with the reordered results. Confirm if high-weight documents such as bid evaluation records are ranked higher. - Simulate a daily incremental update scenario. Check if the data source synchronization is normal, and if the retrieval results include the latest supplier complaint materials or acceptance documents.
The values provided here are common starting points, and should be evaluated against samples relevant to the specific use case.
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.