What this category's data looks like
This category's data looks like this: The management of formal education in technical colleges involves three types of materials: student registration records, teaching plans, and skills training records. Student registration records record students' basic information, registration date, and review signature; teaching plans record course setup, credit-hour allocation, and approval signature; skills training records record training items, completed quantity, date, and instructor signature. Materials are updated by semester or academic year, and the version is marked as current by signature and receipt. When entering the reranking and hybrid retrieval function, student registration records provide the basis for student identity and registration status, teaching plans provide the basis for courses and credit hours, and skills training records provide the basis for the training process and results. The fields of the three types of materials overlap, and names, quantities, and dates need to be checked item by item. When versions are not synchronized or signatures or receipts are missing, the basis entries may point to old versions or invalid records.
Configuration and implementation checks
Configuration and implementation checks: This step is performed by the reranking model in the retrieval service, which outputs the sorted candidates and reranking scores. "Basis entries that can be traced back to the original text" are then completed by citation trimming and answer generation. Hybrid retrieval simultaneously runs 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 |
|---|---|---|
| Student registration records | Fields include student name, student ID, registration date, and review signature; the name, student ID, and date in the same record in teaching plans or skills training records must be consistent | Check whether the name, student ID, and date in student registration records are consistent with those in teaching plans and skills training records; check whether signatures and receipts are complete; confirm that the version is current |
| Teaching plans | Fields include course name, credit hours, and approval signature; the name, credit hours, and date of the same course in the two materials must be consistent | Check whether the course name, credit hours, and date in teaching plans are consistent with those in student registration records and skills training records; check the approval signature and version synchronization status |
| Skills training records | Fields include training item, completed quantity, date, and instructor signature; the name, quantity, and date of the same item in the two materials must be consistent | Check whether the item name, quantity, and date in skills training records are consistent with those in student registration records and teaching plans; check whether signatures and receipts are complete |
| Hybrid retrieval and reranking | Both the vector and full-text paths are enabled and merged, and the reranking model is bound and used for sorting and filtering | In the search parameters, check the two path switches for searchMode; see whether candidates come from both paths at the same time; compare the sorting and scores before and after reranking, use usingReRank to confirm that reranking is in effect, and use limit and similarity to check the number of candidates and the scores |
Three common errors
- Symptom: The name, quantity, and date of the same record do not match between student registration records and teaching plans. Cause: When entering the two sets of data, they were not checked against the same field definitions, or the versions were not synchronized.
- Symptom: The basis entries point to an old version of a teaching plan or skills training record. Cause: The versions were not synchronized, the current version did not replace the old version, and signatures or receipts are missing.
- Symptom: There are many candidates, but the top few are irrelevant, and the results are biased toward keyword hits. Cause: After hybrid retrieval merging, the scope is large, and reranking score filtering was not used, or similarity filtering was not performed according to the reranking score.
How to verify the setup
How to verify the setup: Acceptance for this function: compare the order and scores of the top three before and after enabling reranking, check whether usingReRank is in effect, use limit and similarity to record the number of candidates and the scores, and confirm that the top three after reranking correspond to the basis entries.
Business test: Ask a basis question that spans materials, and see whether the answer gives traceable sources, and whether the sources can be traced back to the original entries in student registration records, teaching plans, or skills training records.
- In the search parameters, check the two path switches for
searchMode, confirm that both the vector and full-text paths are enabled and merged, and check whether candidates come from both paths at the same time. - Compare the sorting and scores before and after reranking, use
usingReRankto confirm that the reranking model is bound and used for sorting and filtering, and check the number of candidates according to the reranking score.