What Data for This Category Looks Like
Data sources for this category include software asset management systems, procurement systems, terminal scanning tools, and vendor authorization documents. Daily documents include license ledgers, software usage lists, procurement contracts, and compliance audit reports. Data is updated monthly or quarterly. Fields include software name, authorized quantity, authorization type, expiration date, number of installed terminals, and others. Units include units, pieces, calendar days, and others. The data structure primarily uses structured ledgers, supplemented by unstructured contracts and audit documents. Field density varies significantly across different documents.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall
The data sources for this category are scattered, and include both structured ledgers and unstructured contracts and audit documents. This requires the retrieval link to adapt to recall rules for both data types. The monthly or quarterly update frequency means high-frequency real-time synchronization is unnecessary. A scheduled incremental update recall trigger logic must be configured instead. Structured data with multiple fields requires retrieval to support precise matching of core fields such as authorized quantity and expiration date. Unstructured documents have large differences in length. Text processing rules for recall must be adjusted to avoid truncating key compliance clauses. The core business requirements of authorization verification and over-installation identification also apply. Recall results must prioritize associating software authorizations with actual installation relationships. This prevents irrelevant content from interfering with judgment.
How to Configure Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category requires precise matching of core fields such as authorized quantity and expiration date. This range filters low-correlation results while retaining weakly associated documents related to compliance |
limit | Top 10 | The core data volume of this category is moderate. 10 results can cover multiple document types including authorization ledgers, procurement contracts, and audit reports, avoiding redundancy |
searchMode | vector + keyword | Balances precise matching of structured fields and semantic recall of unstructured documents, adapting to the mixed data type of this category |
usingReRank | Enabled | Content related to compliance in this category requires strict alignment. Reranking optimizes the relevance ranking of results, preventing core authorization information from being overshadowed by low-correlation results |
datasetSearchUsingExtensionQuery | Enabled | Can cover synonyms or abbreviations through extended queries, adapting to multiple naming scenarios for software names, such as naming differences for the same software from different vendors |
collectionIds | Configured by business scenario grouping | Data in this category is grouped into authorization ledgers, procurement contracts, audit reports, and others. Specifying collection IDs limits the retrieval scope and improves accuracy |
Three Common Misconfigurations
- Symptom: Retrieval results mix data for other software outside this category. Cause:
collectionIdswas not configured to limit the retrieval collection by business scenario, leading to cross-dataset recall of irrelevant content. - Symptom: Responses include general software knowledge outside the knowledge base. Cause: The configuration to strictly generate responses based on recall results was not enabled, leading the large model to call external knowledge.
- Symptom: When no matching authorization-related content is found, a non-preset general response is output. Cause: No custom return logic was configured for when no retrieval results are found, and the retrieval switch that strictly limits knowledge base content was not enabled.
How to Confirm the Configuration Is Correct
- Upload the license ledger and procurement contract for this category, submit a query request for authorized quantity, and check whether the recall results only include document data from the target category.
- Submit a question with no corresponding knowledge base content, and verify whether the preset custom response content is returned.
- Adjust the
similaritythreshold, observe changes in the relevance of recall results, and confirm that it meets the business requirements for accuracy. - Attempt to retrieve documents not configured in
collectionIds, and confirm that irrelevant cross-category data cannot be recalled.
The values provided are common starting points and should be measured against samples relevant to the specific deployment.
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.