What this category's data looks like
This page focuses on compulsory education dropout prevention and school retention materials, sourced from three types of ledgers: dropout risk ledgers, persuasion-to-return records, and home visit records, updated weekly and monthly. Ledger fields include student name, student ID number, risk level, persuasion-to-return date, home visit date, signature and receipt status. When materials enter the knowledge base retrieval service, dropout risk ledgers handle the positioning of risk subjects and levels, persuasion-to-return records handle the positioning of the persuasion-to-return process and results, and home visit records handle the positioning of family communication and signature receipts. The three types of materials serve as candidate fragment sources in retrieval, used to verify whether the name, quantity, and date of the same record match across two materials, and to check whether versions are synchronized and whether signatures or receipts are missing. Progress tables and statistical scope descriptions are generated by subsequent nodes or manually based on this; the retrieval output itself is not a statistical table.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate fragments and similarity scores (after pattern, merge, rerank, filter, and limit trimming). Progress tables and statistical scope descriptions are then completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Dropout risk ledger | Whether fields, dates, and signatures are consistent with persuasion-to-return records | Switch searchMode in the knowledge base search test, and verify the student name, risk level, and date in candidate fragments |
| Persuasion-to-return record | Whether the persuasion-to-return date, persuasion-to-return result, and receipt match the home visit record | Compare the name, quantity, and date of the same student across the two materials, and record the number of candidate items and similarity scores |
| Home visit record | Whether the home visit date, signature, and receipt are complete | Check the similarity scores of returned candidates, and compare the ranking of the top few items before and after enabling usingReRank |
| Statistical scope description | Whether the scope is written in the header and covers exceptions | Use one week of records to verify the total, check whether the scope and exceptions are clearly stated, and separately record the number of cited items |
Three common errors
- Phenomenon: The name or date of the same student does not match across two materials. Cause: The scope was not unified during entry, and versions were not synchronized, causing candidate fragments to point to different records.
- Phenomenon: The quantity of persuasion-to-return records and home visit records is inconsistent. Cause: Receipts or signatures are missing, one material was not updated, and retrieval candidates cannot cover the complete record.
- Phenomenon: There are many candidates, but the ones ranked at the front are all irrelevant. Cause: The retrieval mode does not match the material form, or rerank was not enabled, resulting in the absence of secondary ranking.
How to verify the setup
Feature acceptance: Test once each with synonym-rewritten questions, compare candidates and scores, then toggle rerank to observe the ranking of the top few items; switch searchMode to compare semantic, full-text, and hybrid candidates, adjust similarity and limit to observe filtering and trimming, record the number of cited items separately by array length, and verify the tokens of the cited text separately.
Business test: Use one week of records to verify the total, and check whether the scope and exceptions are clearly stated.
- In the knowledge base search test, switch
searchModeto compare the number of candidates and similarity scores across semantic, full-text, and hybrid modes. - Compare the ranking of the top few items before and after enabling
usingReRank, adjustsimilarityandlimitto observe filtering and citation limit trimming, record the number of cited items separately by array length, and verify the tokens of the cited text separately.