What This Category's Data Looks Like
The business data for this category comes from official operational systems including the national public resource trading platform, natural resource trading systems, and state-owned asset supervision systems. The update frequency is daily. Commonly processed documents include transfer announcements, trading guidelines, bid application forms, margin payment vouchers, transaction confirmation documents, listing result announcements, and property right transfer descriptions. Most documents contain fields such as subject property right numbers, transaction amounts, bid rules, and property right restriction conditions. Some cross-regional or cross-system documents have differences in format and annotation details.
What Constraints Do These Characteristics Impose on the "Knowledge Base Retrieval and Recall" Link
Differences in document formats from multiple systems require retrieval logic to support non-standardized field structures, avoiding missing key property right or transaction information due to format mismatches. Daily updated business data requires the retrieval pipeline to have real-time response capabilities, ensuring that knowledge base data used in the review process is synchronized with current transaction progress. Core verification points vary significantly across different document types. For example, transfer announcements focus on property right restrictions, while bid applications focus on qualification requirements. Retrieval logic must accurately match keywords and fields for the corresponding scenario. Additionally, review errors can directly lead to consequences such as invalid transactions and disputes over state-owned asset losses, requiring recalled results to have high confidence to avoid missing key content.
Reference only; verify against the target FastGPT release and deployment documentation.
How to Set the Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets high confidence requirements for property right review, filters low-relevance distracting results |
limit | Top 10 entries | Covers retrieval needs for multiple associated documents in declaration materials, avoids missing key supporting materials |
searchMode | Semantic matching | Adapts to field and content matching for multi-format documents, supports document structures exported from different business systems |
usingReRank | Enabled | Improves result ranking accuracy in long document scenarios, accurately locates core content such as property right restrictions and transaction rules |
datasetSearchUsingExtensionQuery | Enabled | Supplements contextual information for associated documents, such as recalling corresponding transfer announcements via bid applications |
collectionIds | Pass the exclusive collection ID for the corresponding category | Isolates knowledge base data from other categories, avoids distracting results from unrelated documents |
Three Common Mistakes
- Issue: No progress feedback is shown for batch re-embedding tasks after switching embedding models. Cause: The official FastGPT batch embedding interface was not used, and direct calls to single-document interfaces led to insufficient scheduling resources.
- Issue: No content is returned after adding documents of the corresponding category from the Feishu data source. Cause: FastGPT was not authorized to access the corresponding Feishu document folder, or the public access permission for the documents was not enabled.
- Issue: Empty fields are returned after uploading files to the knowledge base via external interfaces. Cause: A valid API key was not properly carried, or cross-domain access permissions were not configured, resulting in invalid interface calls.
How to Confirm the Configuration Is Correct
- Upload a typical transfer announcement document for this category, trigger the retrieval process, and verify that the similarity scores of the returned results fall within the preset range.
- Check the number of retrieved results, confirm that it matches the value set in the
limitparameter. - After enabling the
usingReRankfunction, compare the ranking of retrieval results with that when the function is disabled, and confirm that the re-ranking logic is active. - Call the batch embedding interface to process a batch of documents for this category, confirm that the task status updates normally with no error messages.
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.