What the data for this category looks like
This category of data mainly comes from project document libraries, configuration lists, meeting minutes, and test reports. It is updated only when a project phase ends or the overall project is delivered. Daily processed documents include completion documents, configuration manuals, training records, and acceptance certificates. Each document usually contains fields such as project association number, version identifier, signature status, and acceptance node. Document formats are mainly PDF, Word, and PPT. The size of a single document is usually within 100 MB. Overall data accumulates gradually as the project progresses.
Constraints on Knowledge Base Retrieval and Recall
Data updates are not real-time. They only sync when a phase or project ends. High-frequency incremental indexing is not required. A scheduled full sync strategy can be used. Documents contain exclusive fields such as project association numbers and version identifiers. Retrieval must prioritize matching project-specific keywords to avoid interference from cross-project documents. Document formats are diverse and single-document content is professional. Multi-format parsing must be supported, and field-level associations must be retained. Overall data accumulates gradually as the project progresses. Retrieval must support limiting recall scope by project group to prevent unrelated documents from mixing into results.
Configuration Settings
The following configuration items are adapted for the knowledge base retrieval and recall module of FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches project-specific fields and technical terms, avoids recalling irrelevant general documents |
limit | top 8–12 entries | The number of documents per project is usually limited. Too many results will increase context redundancy |
searchMode | vector + keyword | Balances semantic matching for technical terms and exact matching for structured fields such as project numbers |
usingReRank | enabled | Semantic similarity of professional documents is easily disrupted by formatting. Reranking improves result relevance |
datasetSearchUsingExtensionQuery | enabled | Adds extended retrieval for exclusive fields such as project numbers and version numbers, covering users' implicit query needs |
collectionIds | specify collections by project | Limits recall scope to only the document collection of the currently handed over project, avoiding cross-project interference |
Three Common Configuration Mistakes
- Recall results include documents not from the current project:
collectionIdsis not configured to limit the collection scope, leading to recall of all documents in the full knowledge base. - Retrieval results cannot match keywords related to project numbers:
datasetSearchUsingExtensionQueryis not enabled, and the retrieval capability for structured fields is not extended. This prevents recognition of user implicit queries for project identifiers. - Results are invalid due to insufficient matching of technical terms:
usingReRankis not enabled, and thesimilaritythreshold is set too low. Low-relevance general documents cannot be filtered out.
How to Verify Successful Configuration
- Manually enter the exclusive number of the current project as a search term. Check if the recall results only include documents from the currently handed over project.
- Enter a combination of technical terms and general terms. Verify that the relevance ranking of retrieval results meets expectations.
- View the knowledge base synchronization logs. Confirm that incremental or full synchronization tasks are only triggered when a project phase ends.
- Check the configuration panel. Confirm that both
usingReRankanddatasetSearchUsingExtensionQueryare enabled.
The values provided here are common starting points. They should be evaluated 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.