What This Category's Data Looks Like
Data for this category originates from social insurance handling systems, personal benefit record systems, labor relations and work injury identification systems, and electronic social security card platforms. The system updates this data daily. Daily documents include benefit application forms, payment period certificates, retirement approval forms, work injury identification materials, and unemployment insurance benefit application materials. Most documents are either structured tables or standardized texts. They include fields such as payment period, benefit calculation base, and eligibility determination conditions. Units include months, years, and multiples of payment bases.
Constraints on Knowledge Base Retrieval and Recall
Daily data updates require the retrieval pipeline to support incremental synchronization. This prevents performance loss from full re-scanning. The mixed structure of structured documents and standardized texts requires the retrieval link to balance keyword matching and semantic understanding to accurately locate core fields such as payment period and benefit standards. Data from multiple sources may have differences in field naming. The retrieval link must uniformly map these fields to standard values to avoid deviations in recall results. This data directly links to benefit payments, so recall results require higher confidence than standard scenarios. The corresponding relationship between policy original texts and business data must be strictly matched.
Configuration Recommendations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches the core correlation between policy original texts and business data, avoids recalling irrelevant documents |
limit | Top 6 results | Controls context length, adapts to multi-document associated query requirements for benefit verification |
searchMode | hybrid | Balances keyword matching and semantic recall, adapts to mixed structured and unstructured document structures |
usingReRank | Enabled | Performs secondary ranking on initial recall results, improves accuracy of core field matching |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonymous expressions related to policies, covers question methods across different handling scenarios |
Three Common Configuration Mistakes
- An error prompt
undefined model must match "^(textappears during the embedding step. The cause is that a text embedding model supported by FastGPT was not selected, and an incompatible model was used incorrectly. - After uploading an Excel format benefit application material, the structured field content in the table is missing from the retrieval results. The cause is that the structured document parsing function of FastGPT was not enabled, and table field extraction rules were not configured.
- The policy original text reference link from the knowledge base is forcibly displayed in the conversation interface and cannot be hidden. The cause is that the "display knowledge base source" switch was not turned off in the conversation configuration, as this function is enabled by default.
How to Confirm Proper Configuration
- Upload an Excel benefit application form to be verified. Check if the parsed text blocks include core fields such as payment period and benefit calculation base.
- Initiate a test query of "What are the policy basis for work injury benefit verification". Check if the number of recall results matches the configured
limitvalue. - View the embedding model selection interface. Confirm that a text embedding model supported by FastGPT is being used, and there are no errors related to
undefined model. - Confirm that the current FastGPT version is v4.17.0. This ensures configuration items are compatible with the version.
The values listed above 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 2026-09-28.