What this category of data looks like
The data sources for this category are structured ledgers exported from internal project management systems, electronic documents of internal regulation compilations, and project acceptance and approval documents. Updates are synchronized with project initiation, progress changes, and regulation revisions, with a maximum update cycle of 7 days. Documents take individual project data as the basic unit, including fields such as project number, affiliated business department, current progress node, budget amount, and associated regulation clause number. Some documents include structured tables and attachment links. Field units include natural days, ten thousand yuan, and others. Single documents have a wide range of character counts, from hundreds to thousands of characters.
Constraints These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
This category of data has both structured fields and non-regulation text, and has strong associated attributes. This requires the retrieval link to cover both field matching and semantic association to avoid retrieving irrelevant content. Data updates do not follow a fixed cycle, so an automatic synchronization mechanism must be configured to ensure the timeliness of retrieved data. Single documents have a wide range of character counts, so chunk processing must adapt to different information densities to avoid incorrect splitting of key fields. Project data is strongly bound to internal regulations, so retrieval must link associated fields to ensure that retrieved regulation clauses match project scenarios. Some documents include attachment links, so the retrieval link must retain associated information for subsequent calls.
How to Configure the Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the semantic relevance between project data and regulation texts, covering the accuracy requirements of most internal scenarios |
limit | Top 8 entries | Adapts to the conventional number of project-associated regulation clauses, avoiding context overload that affects subsequent generation |
searchMode | hybrid | Combines keyword and semantic retrieval, balancing the retrieval effect of structured fields and non-text content |
usingReRank | Enabled | Re-ranks initial retrieval results to improve the display priority of highly relevant content |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval terms based on user questions, covering structured fields such as project numbers and regulation clause numbers |
collectionIds | Collection ID of the target project knowledge base | Limits the retrieval scope to avoid mixing irrelevant knowledge base content |
Three Common Mistakes
- Failing to configure knowledge base retrieval results as judgment conditions in the judgment component. The cause is that the variable mapping switch for retrieval results is not enabled, so the structured data returned by retrieval is not exposed as callable judgment fields.
- Accidentally introducing project knowledge base content when calling a regular conversation node. The cause is that the automatic knowledge base association configuration in the conversation node is not turned off, or the global retrieval scope is not limited to the target project knowledge base.
- Retrieval results not including image links or attachment information within the document. The cause is that attachment metadata was not extracted synchronously during knowledge base import, or no retrieval rule for attachment association was configured in the retrieval link.
How to Confirm the Configuration Is Complete
- Initiate a query that includes a project number, and check whether the retrieval results include the associated regulation clauses of the corresponding project.
- View the retrieval log to confirm that the number of retrieval results meets the requirements of the configured
limitparameter. - Check whether the globally configured
collectionIdsonly include the collection IDs of the target project knowledge base, to avoid mixing irrelevant content. - Initiate a query that includes attachment keywords, and check whether the retrieval results return the corresponding attachment link information.
The values provided are common starting points and should be evaluated using samples specific to the deployment environment.
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.