What this category's data looks like
This page addresses the compulsory education dropout prevention and school retention scenario. The materials come from dropout risk ledgers, persuasion-to-return records, home visit records, and multilingual parallel texts. The ledger contains student ID, risk level, date, and responsible person's signature; the persuasion-to-return record contains persuasion time, method, and receipt signature; the home visit record contains home visit date, participants, parent signature, and receipt. The update cadence is weekly aggregation and monthly verification, and multilingual versions are updated in sync with the ledger. When these three types of materials enter knowledge base retrieval and recall, the ledger plays the role of structured field anchor, the persuasion-to-return record and home visit record play the role of process evidence, and the multilingual parallel text plays the role of terminology consistency reference. The field characteristic is that the name, quantity, and date of the same record in the ledger, persuasion-to-return record, and home visit record must cross-reference each other, and signatures and receipts cannot be missing.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate chunks and similarity scores (after pattern, merge, rerank, filter, and limit truncation). "Unified terminology glossary and parallel text" is completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Dropout risk ledger | Fields, dates, and signatures are complete; the name and quantity of the same record are consistent between the ledger and the persuasion-to-return record | In the knowledge base search test, search by student ID and check whether the candidate chunks cover the ledger fields and signature column; switch searchMode to compare the differences among semantic, full-text, and hybrid recall, and record the number of candidates and scores |
| Persuasion-to-return record | The persuasion time, method, receipt signature, and ledger date match, and versions are synchronized | Use synonymous questions to retrieve the persuasion-to-return record and check whether the returned candidates contain the receipt and signature; compare the ranking of the top few items before and after enabling usingReRank, and check the similarity score and the number of candidates |
| Home visit record | The home visit date, participants, parent signature, and receipt are complete, and the name of the same record is consistent between the ledger and the home visit record | Retrieve the home visit record and check whether the candidate chunks contain the date, participants, and signature; use limit to adjust the number of returned items, and observe and record the tokens of the cited text and the number of array items separately |
| Multilingual parallel text | The same term is written consistently throughout, the translated name can refer back to the original text, and places where versions are not synchronized are marked | Retrieve the parallel text by term and check the correspondence between the translated name and the original text in the candidate chunks; compare the similarity scores and the number of candidates under different question formulations |
Three common errors
- Phenomenon: The name, quantity, and date of the same student do not match between the ledger and the persuasion-to-return record. Cause: The two materials were entered at different times, and cross-reference checking and version synchronization were not performed.
- Phenomenon: The home visit record lacks a parent signature or receipt, and the persuasion-to-return record lacks a receipt. Cause: No signature verification was set up in the material circulation process, and missing items were not checked during archiving.
- Phenomenon: There are many candidates, but the ones ranked at the front are all irrelevant, and the answer lacks key paragraphs. Cause: The retrieval mode does not match the material form, rerank was not enabled, or the similarity filter threshold is too loose.
How to verify the setup
Acceptance of this feature: Test once each with synonym-rewritten questions, compare candidates and scores, then toggle rerank on and off to observe the ranking of the top few items; check the number of returned candidates and scores corresponding to the four identifiers searchMode, similarity, limit, and usingReRank, and record the tokens of the cited text and the number of array items separately.
Business test: Sample three terms and check whether their writing is consistent throughout, and verify whether the translated names can refer back to the original text.
- In the knowledge base search test, switch
searchMode, record the number of candidates andsimilarityscores under semantic, full-text, and hybrid recall respectively, and check whether multi-route merging includes both semantic and full-text results. - Retrieve once each with
usingReRankenabled and disabled, and compare the ranking changes of the top few items; uselimitto adjust the citation limit, check the cited text tokens and the number of array items, and inspect whether the signature and receipt fields are retained in the candidate chunks.