Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
This category of data comes from state-owned assets supervision information systems, enterprise asset management systems, property rights exchange systems, evaluation agency reports and financial statements. Daily documents include property rights registration forms, asset evaluation reports, property rights transfer plans, on-site transaction announcements, and asset disposal approval forms. Data is updated monthly. Most documents contain structured fields such as ownership number, asset category, book original value, transaction base price, etc. Some reports include unstructured evaluation descriptions and transaction process descriptions.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Monthly updated data must be regularly synchronized to the knowledge base to avoid recalling expired asset disposal or property rights registration information, and ensure the timeliness of retrieval results. Mixed document types of structured and unstructured content require adaptive chunking rules to ensure retrieval accuracy for structured fields and semantic matching for unstructured content. Fields such as ownership number and transaction base price in property rights documents are core retrieval terms, and priority matching is required to ensure business relevance of results. Data from different sources has large format differences, so a unified preprocessing process must be implemented to ensure embedding consistency and avoid retrieval deviations caused by format issues. Recall results for state-owned assets-related documents must strictly match business scenarios. The number of recalled entries must be controlled to reduce token consumption, and avoid mixing irrelevant content.
How to Set Configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches core fields of state-owned assets and property rights documents, filters low-relevance casual content, and avoids invalid recalls |
limit | Top 8 entries | Controls the number of single-recall entries to reduce token consumption, and adapts to the medium-scale knowledge base updated monthly for this category |
searchMode | hybrid | Combines keyword and semantic retrieval, adapts to mixed documents with structured fields and unstructured descriptions |
usingReRank | Enabled | Reranks recall results, prioritizes matching core business fields such as asset ownership and transaction base price |
datasetSearchUsingExtensionQuery | Disabled | Data sources for this category are fixed, no additional query extension is required, to avoid introducing irrelevant non-state-owned assets content |
collectionIds | Exclusive dataset ID bound to business | Only retrieves exclusive state-owned assets and property rights documents for this category, avoiding interference from cross-category data |
Three Common Mistakes
- Phenomenon: When the knowledge base retrieval returns no results, automatic online search is triggered, introducing non-state-owned assets related content. Cause: No branch logic after empty results is configured, or all empty result scenarios are incorrectly bound to online search, without distinguishing business scenarios.
- Phenomenon: The number of recalled documents in a single request is too high, causing token consumption to exceed limits. Cause: No reasonable value for the
limitparameter is set, and the number of recalled entries is not controlled based on the monthly updated scale of the knowledge base for this category. - Phenomenon: The generated reply contains unlabeled knowledge base citation information, or irrelevant retrieval metadata is retained. Cause: No citation filtering rules for retrieval results are configured, the prompt does not clearly require only valid business document content to be used, and non-business fields generated during retrieval are not removed.
How to Confirm the Configuration is Correct
- Submit a test question related to property rights registration, check whether the document source of the recall result is the configured exclusive dataset, and confirm that the
collectionIdsconfiguration takes effect. - Enter an irrelevant casual test question, check whether the number of recall results matches the
limitsetting, and confirm that thesimilaritythreshold filters out low-relevance content effectively. - Trigger the test scenario where the knowledge base is empty, check whether the preset branch logic is triggered, and confirm that the empty result processing configuration is correct.
- View the generated reply content, confirm that no unlabeled retrieval metadata is included, and confirm that the citation filtering configuration takes effect.
The values provided are common starting points, and should be evaluated 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.