What the data for this category looks like
Data sources include compliance documents issued by state-owned asset supervision departments, internal company asset ledgers, employee attendance and welfare systems, investment operation weekly reports, and internal office process manuals. Two update schedules apply: fixed-cycle institutional documents are updated quarterly or annually; asset ledgers are updated in real time as property rights change; temporary supervision notifications are synchronized immediately upon release. Most document structures are structured ledgers (with fields such as asset number, book value, storage location), long-text policy documents, and structured form templates. Fields include asset code, assessed value, service life, and approval process nodes. Units commonly used in state-owned asset statistics include ten thousand yuan, square meters, units, and others.
What constraints these characteristics impose on the knowledge base retrieval and recall workflow
Structured asset ledgers have clear dedicated fields, requiring retrieval to precisely match specified fields to avoid vague recall of irrelevant content. Real-time updated asset data requires recall logic to prioritize the latest document versions, preventing the return of expired asset information. Long-text policy documents contain large compliance clauses, requiring proper splitting to avoid context loss. Mixed-type documents (ledgers, policies, forms) require targeted retrieval rule configuration to prevent unbalanced retrieval weights across document types, which would reduce result accuracy.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | State-owned asset management data has high accuracy requirements. Low-match irrelevant results must be filtered to avoid confusion between asset ledgers and policy documents |
limit | 8–12 results | Internal office scenarios require employees to quickly locate core information. Too many results increase screening costs, while too few may miss key asset or policy content |
searchMode | Field matching priority | State-owned asset management companies have a large number of structured asset ledgers. Field matching can accurately recall specified asset information, which performs better than full-text retrieval |
usingReRank | Enabled | When mixed recall of long-text policy documents and structured ledgers occurs, re-ranking optimizes result relevance sorting and improves employee search efficiency |
datasetSearchUsingExtensionQuery | Enabled | Employee questions may omit key fields such as asset numbers. Extended queries can supplement implicit information and improve recall coverage |
collectionIds | Configure dedicated collection IDs grouped by document type | Differentiate collections for asset ledgers, policy documents, office forms, etc., to enable targeted retrieval rule configuration and avoid interference from cross-type results |
Three Common Configuration Mistakes
- Symptom: A large number of irrelevant policy documents are returned when searching asset ledgers, and recall results show scattered matching scores in the interface. Cause: Dedicated
collectionIdsare not configured by document type, and retrieval weights for different document types are not differentiated, leading to interference from cross-type results. - Symptom: Model-related errors are triggered during knowledge base retrieval, and logs show "no valid retrieval context". Cause:
datasetSearchUsingExtensionQueryis not enabled. Employee questions do not carry key fields such as asset numbers, so valid retrieval conditions cannot be generated, leading to model call failure. - Symptom: When searching long-text policy documents, returned result fragments cannot cover complete compliance clauses. Cause:
usingReRankis not enabled, and no relevance sorting is performed on recall results, leading to low-relevance but fully-fragmented results appearing first.
How to Verify Correct Configuration
- Upload a test asset ledger document and a policy document, search using asset numbers and policy keywords respectively, and check whether returned results only include documents of the corresponding type.
- Enter a question that does not carry an asset number, such as "What office assets were added this month", and check whether corresponding ledger data can be recalled to confirm
datasetSearchUsingExtensionQueryis active. - View retrieval logs, check whether the number of returned results is within the range set by the
limitparameter, and low-match results are filtered by thesimilarityparameter. - Upload an expired policy document and the latest version, and check whether the latest version of the document is returned first during retrieval.
The values given 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.