What this category of data looks like
Data sources include the national public resource trading platform, local public resource trading platforms, engineering approval systems, and bidding and tendering supervision systems. Update frequency is daily. Daily documents include bidding announcements, qualification pre-qualification documents, bidding documents, tendering documents, bid evaluation reports, successful bid candidate announcements, and successful bid result announcements. Supporting policy documents and required knowledge content related to party building learning are also included. Most documents are structured or semi-structured text, containing fields such as section information, qualification requirements, bid evaluation rules, and party building knowledge points. There is no unified fixed unit, and content varies based on document type.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The daily updated data source requires an incremental synchronization mechanism. This avoids resource occupation from full repeated synchronization. Mixed documents of multiple types include both structured bidding and tendering fields and unstructured party building text. Balanced adaptability between keyword and semantic retrieval is needed. Documents from different sources have varying formats. A unified preprocessing workflow ensures retrieval consistency. The content boundary between party building learning and bidding and tendering business must be clearly defined. Irrelevant content is prevented from interfering with the accuracy of Q&A for specific scenarios.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Balances retrieval accuracy for party building knowledge points and bidding and tendering business content, and avoids low-relevance content from mixing into corresponding scenarios |
limit | Top 8 entries | Adapts to the volume of mixed documents, controls context length, and avoids exceeding the context limit of the large model |
searchMode | hybrid | Balances keyword matching and semantic retrieval, adapting to structured bidding and tendering fields and unstructured party building text |
usingReRank | Enabled | Rearranges initial retrieval results, prioritizes matching the core demands of the current scenario, and corrects deviations from single semantic retrieval |
datasetSearchUsingExtensionQuery | Enabled | Generates extended queries based on the original question, covering synonymous substitutions of bidding and tendering terminology and party building expressions, improving retrieval coverage |
collectionIds | Configured separately by business and party building datasets | Isolates cross-category data, avoiding interference between bidding and tendering documents and party building learning content in retrieval results |
Three common configuration errors
- Phenomenon: A large number of irrelevant bidding and tendering business documents are mixed in retrieval results. The proportion of party building knowledge points is less than 30%. Cause:
collectionIdsis not configured to isolate the retrieval scope by dataset, resulting in cross-category data being retrieved simultaneously. - Phenomenon: The number of results returned by a single retrieval is less than 3. It cannot cover complete party building knowledge points or bidding and tendering details. Cause: The
limitparameter is set too low, failing to adapt to the retrieval needs of mixed documents. - Phenomenon: Generated answers have terminology deviations. For example, "qualification pre-qualification" is mistakenly written as "qualification review". Core expressions of party building learning are confused. Cause:
datasetSearchUsingExtensionQueryis not enabled, failing to cover scenario-specific terminology synonymous expression differences.
How to confirm the configuration is correct
- Upload a single bidding and tendering document and a single party building learning document. Initiate a question for the corresponding scenario. Verify that returned results only contain document content of the corresponding category.
- Adjust the
similarityparameter. Verify changes in the number and relevance of retrieval results under different thresholds. Confirm that the value meets business needs. - Enable
usingReRank. Compare the result ranking before and after rearrangement. Confirm that core relevant content is displayed first. - View retrieval logs. Confirm that extended queries generated by
datasetSearchUsingExtensionQuerycover scenario-specific terminology.
The values provided 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.