What data looks like for this category
Data for this category primarily comes from local public data open platforms, integrated big data platforms, departmental business databases, and electronic certificate databases. It covers types such as public data resource catalogs, data opening lists, data item descriptions, desensitization rules, interface documents, and update frequency descriptions. Data exists as a mix of structured fields and unstructured documents, including fixed fields like data codes, opening conditions, interface paths, and desensitization requirements. The update frequency is once per month.
What constraints these characteristics impose on knowledge base retrieval and recall
The monthly update requirement means the retrieval system must align with a fixed index update cycle to avoid retrieving expired data. The mixed structured and unstructured data type requires the retrieval mode to support both field matching and full-text search, covering scenarios such as data item codes and interface document content. Data involves opening authorization and desensitization rules, so retrieval and recall must strictly limit the dataset to the authorized scope to prevent unauthorized access. There are a large number of data entries: retrieving too many at once will exceed the context length, while retrieving too few will fail to cover all required data items.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches the relevance between structured fields and unstructured descriptions of public data catalogs, avoids returning irrelevant data if set too low, and misses matching items if set too high |
limit | Top 10 entries | There are many public data catalog entries. Retrieving too many at once will exceed the context length, while retrieving too few will fail to cover required data items |
searchMode | hybrid | Supports both structured data item retrieval and unstructured document content recall, adapts to mixed-type data such as catalogs and interface documents |
usingReRank | Enabled | Reorders retrieved entries by relevance, optimizes sorting results for long texts such as interface documents and desensitization rules |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms to match fields easily overlooked by users, such as data item aliases and interface paths |
collectionIds | Fill in according to the specified data directory collection | Limits the retrieval scope to only the knowledge base collections related to public data catalogs, avoids mixing in other business data |
Three common mistakes
- When uploading PDF documents for public data catalogs, the interface displays "Unable to read the file content". The cause is that the document contains encrypted content or uses a non-standard PDF format. Public data catalog PDFs are mostly official document formats, and some encrypted versions cannot be chunked and parsed.
- When batch uploading documents related to public data catalogs, a network failure prompt appears or some files fail to upload. The cause is that the total file size of a single upload exceeds the limit, or network fluctuations cause chunked upload interruptions.
- When configuring a Feishu-synced public data catalog knowledge base, PPT and PDF files cannot be synchronized. The cause is that the synchronization permission for the corresponding file types is not enabled, or the file format of the Feishu data source is not supported.
How to confirm configurations are set correctly
- Perform a retrieval targeting data item fields, check if the returned result fields match the structured fields of the public data catalog.
- Upload a test public data interface document, confirm that it can be properly chunked and generate retrieval indexes.
- Adjust the
similarityparameter, verify that the relevance of the returned results meets expectations. - After enabling
usingReRank, compare the sorting differences before and after reranking to confirm the reranking function is active.
The values provided 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.