What this category's data looks like
This page addresses the compulsory education dropout prevention and control scenario. Materials come from two levels, school and class: dropout risk ledgers, persuasion-to-return records, home visit records, and resumption-of-school applications and receipts submitted by parents. The update rhythm is centralized ledger creation at the start of each semester, dynamic supplementary entry each month, and persuasion-to-return and home visits recorded as events occur. Field characteristics center on student name and student registration number, risk level, persuasion-to-return date and result, home visit date and signature, and delivery receipt. When materials enter this function, the ledger handles locating risk subjects, persuasion-to-return records and home visit records provide process and result evidence, and resumption-of-school applications and receipts handle destination confirmation; the vector path recalls semantically similar passages, the full-text path recalls precise fields such as student registration number and dates, and after the two paths are merged, the reranking model uniformly sorts them for subsequent citation trimming and answer generation.
Configuration and implementation checks
This step is executed by the reranking model in the retrieval service, which outputs the sorted candidates and rerank scores. "Evidence items that can be traced back to the original text" are then completed by citation trimming and answer generation. Hybrid retrieval simultaneously runs the vector and full-text paths and merges the results; after merging, the scope is large and similarity filtering cannot be applied directly, so the reranking model is usually used again to sort and filter by rerank score.
| Object | Check | Verification action |
|---|---|---|
| Dropout risk ledger | Fields include student name, student registration number, risk level, ledger creation date, and homeroom teacher signature; the date must be comparable with the first date in the persuasion-to-return records | In the search parameters, check whether searchMode enables both vector and full-text, see whether candidates come from both paths, and record the number of candidates and scores |
| Persuasion-to-return record | Fields include persuasion-to-return date, method, result, and parent signature; the name, quantity, and date of the same record in the two materials must be consistent | Compare the ordering and scores before and after reranking, check that usingReRank is bound and used for sorting and filtering, and confirm whether the top few items fall on persuasion-to-return records |
| Home visit record | Fields include home visit date, visitor, parent signature, and receipt; entries missing a signature or receipt cannot be used as evidence | After checking that limit controls the number of returned items, trace each item back to the original passage and confirm that the home visit date matches the ledger creation date |
| Resumption-of-school application and receipt | Fields include application date, resumption grade, and receipt acknowledgment; the version must be the current one | After checking similarity and rerank score filtering, verify whether the receipt is the same version as the persuasion-to-return result |
Three common errors
- Phenomenon: the dates of the same persuasion-to-return record do not match between the ledger and the home visit record. Cause: the two are entered separately and version synchronization is not performed, so they may still be presented side by side after reranking.
- Phenomenon: there are many candidates but the top few are irrelevant. Cause: after merging, the scope is large and filtering by rerank score is not applied, so keyword hits crowd out semantically relevant items.
- Phenomenon: the home visit record lacks a parent signature or receipt. Cause: supplementary entry of materials is incomplete, so although the item is recalled, it cannot be traced back to valid original text.
How to verify the setup
This function's acceptance: compare the order and rerank scores of the top three items before and after enabling reranking, and check whether the number of candidates and scores change with usingReRank.
Business test: ask a cross-material evidence question and see whether the answer provides a traceable source.
- In the search parameters, check the two path switches of
searchMode, and see whether candidates come from both the vector and full-text paths. - Compare the ordering and scores before and after reranking, trace each item back to the original text, and confirm that dates, quantities, signatures, and receipts match.