Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Core data for real estate registration business comes from the real estate registration information system, cadastral survey database, housing transaction online signing system, tax system, and electronic certificate library. The update frequency is daily.
Daily involved documents include registration application forms, ownership certificates, survey results, tax payment vouchers, real estate ownership certificates, and registers. Document structures cover fields such as subject information, survey parameters, tax details, and right restriction status. There is no unified fixed single-document format: some documents use structured fields, while others use unstructured text.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Multiple and heterogeneous data sources require retrieval to adapt to different document formats, avoiding missing key information due to format differences. The daily update frequency requires retrieval to cover the latest registration data, ensuring the timeliness of business processing. The core business is confirmation of real estate rights, requiring retrieved content to accurately hit core fields such as ownership source, restriction status, and survey comparison, avoiding irrelevant content interfering with the verification process. The consequences of errors are directly linked to registration errors and property rights disputes. Retrieval confidence must meet business compliance requirements, and low-match content must not be introduced.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Match precise content related to real estate ownership, filter low-relevance non-business documents, and meet the high compliance requirements of registration business |
limit | Top 8 entries | Cover necessary information for multiple types of business documents such as ownership source verification, survey comparison, and tax vouchers, avoiding excessive retrieved results causing context overflow |
searchMode | hybrid | Combine full-text retrieval and vector retrieval, adapt to heterogeneous registration business document formats, and balance retrieval effects for structured fields and unstructured text |
usingReRank | Enabled | Reorder retrieved results by relevance, prioritize returning matching content of core business fields such as ownership source and restriction status, and improve the targeting of generated responses |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand relevant business terms based on user questions, associate common expressions such as "transfer" and "mortgage" with documents of corresponding registration types, and improve retrieval accuracy |
collectionIds | Configured for dedicated real estate registration datasets | Limit the retrieval scope to only include dedicated documents such as registration application forms and ownership certificates, avoiding irrelevant data from other categories |
Three Common Mistakes
- Phenomenon: Retrieval hits knowledge base documents, but the generated answer only provides a brief summary without displaying specific clause content. Cause:
datasetSearchUsingExtensionQueryis not enabled, and specific clause field information is not associated, causing the model to fail to obtain complete document content. - Phenomenon: Single retrieval takes more than 30 seconds, and the page returns a timeout status code. Cause:
collectionIdsis not limited to a dedicated dataset, and the retrieval scope covers too many heterogeneous system documents, resulting in excessive vector retrieval calculation load. - Phenomenon: The number of retrieved results is inconsistent each time for the same question. Cause: The
similaritythreshold is not fixed, or the fixed sorting rule ofusingReRankis not enabled, resulting in strong randomness of retrieved results.
How to Confirm the Configuration is Correct
- Initiate a test query containing "ownership source certificate", check whether the retrieved results include registration application form and ownership certificate documents.
- View the retrieval log, confirm that the number of retrieved results matches the configured value of
limit, and the sorting conforms to the reordering rule ofusingReRank. - Test initiating the same query 3 consecutive times, confirm that the number and sorting of retrieved results are consistent.
- Verify that for the dedicated dataset configured with
collectionIds, the retrieval results do not include non-real estate registration document content.
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.