Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Data for this category mainly comes from special equipment safety supervision systems, inspection and testing institution systems, user account books, and operator certificate databases. It covers core documents including use registration certificates, periodic inspection reports, maintenance records, hidden danger rectification notices, accident reports, and operator certificates. The data update frequency is monthly. Core fields include equipment number, inspection cycle, maintenance timeliness, operator qualification number, hidden danger rectification status, and others. Most document structures are structured forms and structured records, with some semi-structured content transcribed from scanned documents.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Monthly updated data requires regular synchronization to the knowledge base to prevent expired data from interfering with the compliance of retrieval results. Precise structured fields such as equipment number and inspection cycle require retrieval to balance exact matching and semantic recall, and cannot rely solely on a single recall logic. Semi-structured content transcribed from scanned documents is prone to transcription errors, requiring more detailed chunking and recall rules to ensure key information is not missed. There is a need for associated retrieval between operator certificates and equipment maintenance records, requiring recall results to cover both the equipment subject and relevant personnel qualification information, avoiding single-dimensional results that cannot support complete business judgments. Multi-source data requires support for cross-dataset joint retrieval to ensure the recall scope covers all compliant data sources.
How to set the configuration
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Balances the need for exact matching of structured fields such as equipment number and inspection cycle with semantic recall of maintenance records, avoiding missing valid content if the threshold is too high, or introducing irrelevant data if the threshold is too low |
limit | Top 8–12 entries | A single recall must cover multiple types of documents for a single device, including registration, inspection, maintenance, rectification, etc., while controlling context length to fit model constraints |
searchMode | hybrid | Supports both exact matching of structured fields and semantic recall of semi-structured documents, adapting to retrieval needs for multiple data types |
usingReRank | Enabled | Semantic similarity of semi-structured transcribed content is easily affected by transcription errors, and reranking can optimize the relevance ranking of recall results |
datasetSearchUsingExtensionQuery | Enabled | In internal office scenarios, employee questions often implicitly include preconditions such as equipment number or cycle, and extended queries can supplement associated information to improve recall accuracy |
collectionIds | Bind dedicated datasets for special equipment supervision | Limit the retrieval scope to dedicated documents related to supervision, avoiding introducing irrelevant data from other categories |
Three common configuration errors
- A 418 status code appears after configuring the knowledge base. Cause: Failed to correctly bind the special equipment supervision dedicated dataset specified by
collectionIds, leading to retrieval scope conflict or permission exception. - Recall results only meet partial conditions when submitting a question. Cause: The
similaritythreshold is set unreasonably, missing associated documents that contain complete business conditions, orusingReRankis not enabled leading to ranking deviation. - Content cannot be obtained when importing WeChat Official Account links. Cause:
datasetSearchUsingExtensionQueryis not enabled to support external link parsing, or the link format does not meet the compliance requirements for knowledge base import.
How to confirm the configuration is correctly set
- Check the dataset binding relationship, confirm that
collectionIdsonly includes dataset IDs related to special equipment supervision. - Initiate a test retrieval, enter a question containing equipment number or inspection cycle, verify whether the recall results cover multiple types of documents such as registration, inspection, and maintenance.
- Adjust the
similaritythreshold, compare the number of recall results under different thresholds to confirm that accuracy and recall scope are balanced. - Import a test external link content, verify that it is successfully synchronized to the specified dataset with no error prompts.
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.