What this category's data looks like
Adult college entrance examination registration and admission management involves three types of core materials: registration forms, preferential scoring materials, and admission rosters. Registration forms record the candidate's basic information, application level, and signature confirmation; preferential scoring materials contain the scoring item, certificate number, and review signature; admission rosters record the admission result, major, and date. The three types of materials are updated by year and batch, and fields include name, ID number, date, quantity, signature, and receipt. When the materials enter the reranking and hybrid retrieval function, the registration forms and scoring materials serve as basis sources, while the admission roster serves as the result verification object; each participates in candidate ranking in the form of original-text entries, for citation trimming and answer generation to refer back to the original text.
Configuration and implementation checks
The reranking model in the retrieval service only outputs the sorted candidates and reranking scores; the basis entries that can refer back to the original text must then be handled by citation trimming and answer generation. Hybrid retrieval runs both the vector and full-text paths and merges the results; after merging, the scope is large and similarity filtering cannot be done directly, so the reranking model is usually then used to sort and filter by reranking score.
| Object | Check | Verification action |
|---|---|---|
| Adult college entrance examination registration form | Fields include name, ID number, application level, date, and signature, and must match the preferential scoring materials and admission roster item by item | Check the two-path switches of searchMode in the search parameters, confirm that both vector and full-text are enabled, and see whether candidates come from both paths at the same time |
| Preferential scoring materials | The scoring item, certificate number, review signature, and date must be consistent with the registration form, and the version must be current | Compare the ranking and scores before and after reranking, check that usingReRank is already used for sorting and filtering, and check whether the top few results under the similarity threshold are relevant |
| Admission roster | The admission result, major, date, and quantity must correspond to the registration form and scoring materials; missing signatures or receipts must be flagged | Use limit to control the number of candidates, record the number of candidates and the reranking scores, and refer back to the original text to check the name, quantity, and date |
Three common errors
- Symptom: the same candidate's name is written differently in the registration form and the admission roster, and the quantity or date does not match. Cause: field alignment was not done when the two materials were entered, and the versions were not synchronized.
- Symptom: preferential scoring materials lack a review signature or receipt but still enter the candidates. Cause: the material version was not synchronized, and the signature and receipt fields were not verified.
- Symptom: there are many candidates but the top few are irrelevant, and similarity filtering appears to have failed. Cause: after merging, the scope is large, and reranking scores were not used for filtering.
How to verify the setup
Acceptance of this function: compare the order and reranking scores of the top three before and after enabling reranking, check usingReRank, similarity, limit, and searchMode, and record the number of candidates and the scores.
Business test: ask a basis question that spans materials, and see whether the answer gives a citable source.
- Check the two-path switches of
searchModein the search parameters, confirm that both vector and full-text are enabled, and see whether candidates come from both paths at the same time. - Compare the ranking and scores before and after reranking, check that
usingReRankis already used for sorting and filtering, and use thesimilaritythreshold andlimitto control the number of candidates and the scores.