What data looks like for this category
Data for this category primarily comes from asset management systems, procurement systems, financial fixed asset modules, and terminal management platforms. Daily generated documents include asset ledgers, inventory sheets, loan and return forms, and scrap applications. Core fields include asset number, name, ownership, storage location, user, status, purchase date, and others. Data update frequency is mostly monthly, quarterly, or annually. Some temporarily loaned or returned assets will have additional temporary updates. Overall data volume fluctuates with the enterprise’s asset scale.
What constraints do these characteristics impose on knowledge base retrieval and recall
Data sources for this category are scattered, requiring integration with multiple systems for data synchronization. As a result, knowledge base retrieval must support unified recall logic across multiple data sources. Data update frequency is low but batch size is large. The recall link must support switching between incremental and full updates to avoid reloading old data. Core fields focus on asset ownership, location, user, and other information. Retrieval must balance exact matching and fuzzy matching needs to ensure accurate recall of key information required for account-physical verification. Asset data accuracy is directly linked to audits and asset security. Recall results must have high confidence levels, filtering expired or invalid asset entries.
How to configure settings
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for core fields such as asset number and user, filtering low-correlation non-target asset data |
limit | Top 8–12 entries | Covers the number of asset entries required for a single work order, avoiding context window overload |
searchMode | Vector + keyword hybrid retrieval | Adapts to the exact matching needs of short text fields such as numbers and names in asset ledgers |
usingReRank | Enabled | Performs secondary sorting on recall results to improve the display priority of core asset information |
datasetSearchUsingExtensionQuery | Enabled | Optimizes query statements by adding synonym extensions related to assets, expanding retrieval recall scope |
collectionIds | Exclusive collection ID for the IT asset inventory knowledge base | Limits retrieval scope to only exclusive data for this category, avoiding mixing content from other business knowledge bases |
Three common configuration errors
- Phenomenon: Expired scrap asset entries appear in retrieval results. Cause: No filtering rule based on data update time is configured, recalling invalid archived historical data.
- Phenomenon: Asset ledger data can be queried in the knowledge base backend, but no corresponding content is output after being passed to the large model. Cause: The
datasetSearchUsingExtensionQueryparameter is not enabled, or thelimitvalue is set too low, causing recall results to not enter the context window. - Phenomenon: Matching fails for the user field on loan and return forms, only returning some relevant assets. Cause:
searchModeis only configured as vector retrieval, and keyword matching is not enabled, making it impossible to perform exact matching on short text fields such as names.
How to confirm correct configuration
- Upload a latest asset inventory sheet to the test environment, execute a query statement containing an asset number, and verify that the recall results include valid asset entries from the current cycle.
- Call the retrieval interface, check the number of returned results, and confirm that it matches the configured
limitvalue. - Enter a query statement containing a user’s name, and verify that the retrieval results prioritize returning asset information managed by that user.
- Check the knowledge base synchronization logs for FastGPT v4.17.0, confirming that multi-source asset data has completed synchronization according to the configured update frequency.
The values listed above are common starting points. Always test and adjust based on your own specific samples and operational scenarios.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.