What data in this category looks like
Data for this category typically comes from customer ITSM systems, monitoring platforms, asset inventories, and internal on-duty records. Update cadence falls into two categories: daily and monthly.
Commonly produced documents include daily on-site reports, service desk tickets, inspection reports, and SLA monthly reports. Daily on-site reports are updated daily, and include on-site issue descriptions, processing progress, and responsible parties. Service desk tickets include ticket numbers and status fields. Inspection reports summarize device operating parameters and check items on a monthly basis. SLA monthly reports include compliance rate and penalty-related fields. Document structure is mostly semi-structured, mixing structured fields and free-text descriptions.
What constraints these characteristics impose on knowledge base retrieval and recall
Data sources are scattered across multiple customer-side IT systems and internal records. Retrieval configurations must support multi-collection association to avoid missing core business documents in the retrieval scope. Update frequencies are split into daily and monthly categories. Incremental update trigger rules must be configured to reduce resource consumption from full retrievals. Documents include structured fields such as ticket numbers and SLA compliance rates. Field-level precise retrieval must be supported to improve recall efficiency for specific information. Documents in on-site scenarios are mostly free-text records created on the spot. Balance must be struck between semantic matching precision and recall coverage. The total length of single recall results must also be limited to fit large model context limits.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Fits semantic matching needs for semi-structured documents such as on-site tickets and inspection reports, balancing precision and recall coverage |
limit | Top 8 entries | Fits scenarios where a single on-site service retrieval needs to cover multiple tickets or daily reports, avoiding large model context overload |
searchMode | hybrid | Combines keyword matching and semantic retrieval, adapting to documents that include structured fields such as ticket numbers and free text |
usingReRank | Enabled | Performs secondary ranking on recall results, corrects matching deviations for semi-structured documents, and improves recall accuracy for SLA-related content |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands user queries, supplements common operational terminology for on-site scenarios, and optimizes retrieval recall scope |
collectionIds | Associate target collections such as customer ITSM and inspection reports | Limits retrieval scope to only business documents associated with the current on-site service, avoiding interference from irrelevant data |
Three common errors
- Symptom: Retrieval results return empty or far fewer entries than expected. Cause: The configured
similaritythreshold is too high, filtering semi-matching content suitable for on-site scenarios, such as inspection reports that only match partial keywords. - Symptom: Locally stored knowledge base images fail to display properly. Cause: The association logic for permanent image storage is not configured correctly. Temporary storage paths are mistakenly bound to retrieval tasks, causing images to expire along with the temporary storage period.
- Symptom: Ticket data from customer ITSM systems cannot be retrieved. Cause: The
collectionIdsconfiguration is not set to associate the knowledge base collection of the corresponding data source, resulting in the retrieval scope not covering the target data source.
How to confirm configurations are set properly
- Access the FastGPT v4.17.0 knowledge base management interface, and check if the associated
collectionIdscover business data source collections such as customer ITSM and inspection reports. - Initiate a test retrieval, input "Last week's on-site ticket processing status", and verify that the number of returned results matches the
limitsetting. - Trigger an incremental sync, check if the data update frequency aligns with the daily and monthly business cadence, and confirm there are no abnormal error logs.
- After enabling
usingReRank, compare the result rankings before and after reranking, and confirm that SLA-related content ranks higher.
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 2026-10-19.