What this category of data looks like
This category of data typically originates from integrated budget management systems, departmental staff establishment systems, prior-year final accounts, project repositories, and budget preparation notices issued by the Ministry of Finance. Commonly processed documents include departmental budget drafts, project application forms, performance target forms, personnel expense calculation forms, and budget approval documents.
Data updates follow an annual cycle. Most documents are a mix of structured tables and standardized text. Fields include budget subject codes, staff establishment counts, project performance indicators, expense calculation standards, and other items. There are no uniform fixed units. Matching the measurement rules of corresponding policy documents is required.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The annual update requirement means the knowledge base only needs to sync data in bulk once per year, without frequent incremental updates. However, the high total volume of documents imported in a single batch and their complex field structure increase indexing time.
The mixed structure of structured tables and standardized text requires retrieval to adapt to splitting and association rules for different content types. This prevents cross-page budget subjects and policy explanations from being split into separate chunks.
Business data and policy documents from multiple sources must be unifiedly linked. Retrieval and recall must match both business fields and regulatory basis. This ensures results cover both business data and policy requirements.
Additionally, results from this link directly affect budget review pass rates. The relevance range of recalled content must be strictly limited to prevent irrelevant entries from interfering with core judgments.
How to set the configuration
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the association strength between policy texts and business data for departmental budget preparation. This range filters low-relevance general policy entries, retaining content directly related to specific budget subjects and expense calculations. |
limit | Top 10 entries | Number of results returned per recall. This fits budget preparation review scenarios. It provides sufficient policy reference without overloading reviewers with excessive entries. |
searchMode | keyword+vector | Combines keyword matching and vector retrieval. This covers both clear policy clause keywords and vague business scenario descriptions. It adapts to the retrieval needs of mixed document types. |
usingReRank | Enabled | Performs reordering on recalled results. Prioritizes entries directly related to budget subjects and expense calculations. This improves the display priority of core content. |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms. Adds synonyms and common phrases related to budget preparation. This covers the query habits of different job roles, such as expanding "personnel expenses" to "personnel public expenses" and "personnel salary calculation". |
collectionIds | Bind dedicated departmental budget preparation dataset | Limits retrieval scope to only the dedicated document collection for this category. Prevents irrelevant data from other fiscal business from being included. This ensures the targeted nature of recalled results. |
Three common mistakes
- An upstream connect error or disconnection prompt appears when uploading Excel files. The cause is that custom separators are not configured correctly. This causes individual chunk content to exceed system processing thresholds, triggering connection interruptions.
- Knowledge base indexing speed is too slow. The cause is that
collectionIdsis not limited to the dedicated dataset. Importing redundant cross-year historical data at the same time increases indexing computation load. - Imported Excel split results merge multiple rows into one entry. The cause is that custom separators are not enabled. The system splits by default line breaks, and does not recognize Excel row-level separation rules. This causes cross-row content to be merged.
How to confirm correct configuration
- Upload test departmental budget draft and policy notice files. Check the chunked results after import. Confirm that each Excel row content is split correctly.
- Initiate a query containing a specific budget subject name. Verify that the
similarityscore range of recalled results matches the preset range. - Call the knowledge base interface to confirm that the number of returned results matches the
limitconfiguration value. Ensure no content from irrelevant datasets is included. - Test result ordering after enabling
usingReRank. Confirm that core policy clauses are displayed first.
Note: The configuration values listed above are common starting points. They should be evaluated against the reader's own sample data.
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.