What the data for this category looks like
This category of data mainly comes from daily work documents of privacy compliance teams, security consulting and legal departments, including personal information lists, processing rules, impact assessment reports, user authorization texts, etc. Data sources cover business systems, App permission lists, privacy policies and user request records. The update frequency is quarterly or each time a product function is changed. Documents include structured table content and unstructured compliance report content. Most fields focus on natural person information identification and processing legality, with no additional non-compliant business fields.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Since the data includes structured tables and unstructured compliance texts, retrieval must support parsing and matching both types of documents. The periodic update requirement means the knowledge base must regularly synchronize the latest versions of compliance documents to avoid using outdated processing rules or authorization texts. Document fields focus on personal information-related content, so the retrieval scope must be strictly limited to the compliance knowledge base to prevent irrelevant business data from mixing into results. At the same time, the accuracy of compliance content is extremely high, so the recalled results must highly match the compliance queries in user tickets to avoid misleading ticket processing by on-call operations staff.
How to set the configuration
The configuration logic on this page is adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75-0.85 | Compliance content requires high relevance to avoid irrelevant documents interfering with compliance judgments for tickets |
limit | Top 6-8 entries | Internal ticket queries require quick access to compliance basis; too many results will exceed the context length limit |
searchMode | hybrid | Simultaneously adapts to retrieval needs for structured tables such as personal information lists and unstructured compliance texts |
usingReRank | Enabled | Re-ranking can improve the sorting priority of highly relevant compliance documents, ensuring key basis is displayed first |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval terms based on user ticket questions to cover more associated compliance document content |
collectionIds | Specify exclusive IDs for compliance knowledge bases | Limit the retrieval scope only to documents related to personal information protection compliance, avoiding mixing in non-compliant data |
Three common mistakes
- Phenomenon: When asked to repeat the content of an uploaded xlsx-format personal information list in a conversation, the system returns that it cannot read the file. Cause:
datasetSearchUsingExtensionQueryis not enabled, and the retrieval and parsing rules for table files are not configured. The system only supports reading text documents by default. - Phenomenon: Retrieval results include non-compliant business knowledge base content, and cannot accurately match documents related to personal information protection. Cause:
collectionIdsis not specified, so the retrieval scope covers other business knowledge bases, leading to mixed results that do not meet compliance data isolation requirements. - Phenomenon: Retrieval returns results with too low relevance, or misses key compliance basis. Cause: The
similaritythreshold is set incorrectly, and no actual calibration is performed based on the text characteristics of compliance documents, resulting in insufficient matching accuracy.
How to confirm the configuration is complete
- Upload a test personal information list xlsx file, initiate a ticket containing "View the content of the 10th line of this file", and check whether the complete content of the corresponding table row can be returned.
- Check the retrieval log to confirm that only documents from the specified compliance knowledge base are matched, and no results from other non-compliant knowledge bases are returned.
- Adjust the
similaritythreshold, verify the change in retrieval result relevance under different thresholds, and confirm that it meets the accuracy requirements for compliance content. - After enabling
usingReRank, compare the result sorting before and after re-ranking to confirm that highly relevant compliance documents are displayed first.
The values given 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 2026-10-19.