What this category's data looks like
In the scheduling and class-hour deduction scenario of training institutions, progress records and statistical reconciliation rely on three types of materials: class schedules, class-hour deduction records, and schedule-change requests. Class schedules are updated weekly or monthly, with fields including course name, class, teacher, classroom, date, and time slot; class-hour deduction records are generated one by one, with fields including student, course, date, number of class hours, and signature; schedule-change requests record information before and after the change, including request date, reason, approval, and receipt. After these three types of materials enter the knowledge base retrieval and recall function, each serves as a source of candidate snippets: class schedule snippets are used to check the scheduling criteria, class-hour deduction record snippets are used to check actual consumption, and schedule-change request snippets are used to check exceptions and adjustments. In terms of update rhythm, class schedules change as scheduling adjustments are made, class-hour deduction records increase as classes take place, and schedule-change requests are supplemented as approvals circulate. In terms of field characteristics, the names, quantities, and dates of the same record may be inconsistent across different materials; when versions are not synchronized or signatures/receipts are missing, the retrieved candidate snippets will show differences, and relevance must be judged in combination with similarity scores.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate snippets and similarity scores (after mode, merge, rerank, filter, and upper-limit trimming). The "progress table and statistical criteria description" is completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Class schedule | Whether the name, date, and time slot are consistent with the class-hour deduction records; whether the signature column is complete | In the search test, switch searchMode, use synonymous questions to retrieve class schedule snippets, record the number of candidates and scores, and check the fields and signatures |
| Class-hour deduction records | Whether the number of class hours and date match the class schedule and schedule-change requests; whether the receipt is missing | Compare the ranking of the first few items before and after enabling usingReRank, and check whether the class-hour deduction record snippets retain the key original fields |
| Schedule-change requests | Whether the name, quantity, and date before and after the change are consistent with the class schedule and class-hour deduction records; whether the version is synchronized | Adjust similarity to observe candidate changes, and check the approval and receipt information in the schedule-change request snippets |
| Statistical criteria description | Whether the header states the correspondence between the number of records and the statistical count and the handling of exceptions | Use one week of records to check the totals, check whether the criteria and exceptions are clearly stated, and record the number of candidates and scores |
Three common errors
- Symptom: The name, quantity, and date of the same record do not match between the class schedule and the class-hour deduction records. Cause: Material versions are not synchronized, or entries were not filled in according to unified criteria.
- Symptom: Schedule-change requests lack signatures or receipts, but class-hour deduction records are still counted according to the original class schedule. Cause: The approval circulation is not closed-loop, and exceptions are not explained in the statistical criteria.
- Symptom: There are many candidates, but the ones ranked first are all irrelevant, and the answer lacks key paragraphs. Cause: The retrieval mode does not match the material form, or rerank is not enabled, causing ranking bias.
How to verify the setup
This function's acceptance: test once each with synonymously rewritten questions, compare candidates and scores, then toggle rerank on and off to see the ranking of the first few items; switch searchMode to observe differences between semantic and full-text results, and adjust similarity and usingReRank to record changes in the number of candidates and scores.
Business test: use one week of records to check the totals, and see whether the criteria and exceptions are clearly stated; compare the three materials—class schedule, class-hour deduction records, and schedule-change requests—item by item for name, quantity, date, and signature.
- In the knowledge base search test, retrieve with different phrasings of the same question, record the number of candidates and similarity scores, and check whether the returned snippets cover the key fields of the class schedule, class-hour deduction records, and schedule-change requests.
- Against one week of class schedules, class-hour deduction records, and schedule-change requests, check item by item for name, quantity, date, and signature/receipt, and write inconsistent items and exception handling into the header of the statistical criteria description table.