What this category's data looks like
The registration review materials for professional and technical personnel vocational qualification examinations include registration forms, education certificates, and work experience certificates. They come from candidates' submissions and employers' issuance, and their update rhythm rolls with batch applications, supplementary documents, and reviews. The registration form contains name, ID number, examination subjects, work unit, signature, and date; the education certificate contains school, major, graduation date, and certificate number; the work experience certificate contains start and end dates, position, employer seal, and handler's signature. When the materials enter the knowledge base retrieval service, each participates in recall as an independent fragment: the registration form provides identity and examination application fields, the education certificate provides education and time fields, and the work experience certificate provides years and signature fields. The retrieval stage is only responsible for finding candidate fragments and similarity scores related to exception tracing, for subsequent nodes or manual comparison of whether fields, dates, and signatures are consistent.
Configuration and implementation checks
The knowledge base retrieval service (retrieval node and search test) only outputs candidate fragments and similarity scores (after mode, merge, rerank, filter, and upper limit trimming). The exception list and review conclusions must be handled by subsequent nodes or manually. One retrieval consists of question optimization, semantic/full-text/hybrid recall, multi-route merge, rerank secondary sorting, similarity filtering, and citation upper limit trimming, and only then is the citation content handed to the model obtained.
| Object | Check | Verification action |
|---|---|---|
| Registration form | Whether fields, dates, and signature match the education certificate and work experience certificate | In the search test, switch searchMode, and for the same exception record use semantic, full-text, and hybrid recall respectively, record the number of candidate items and scores, and check whether the registration form's name, ID number, examination subjects, and signature date are consistent |
| Education certificate | Whether school, major, and graduation date match the registration form and work experience certificate | Compare the top few rankings before and after enabling usingReRank, see whether the education certificate's school, major, and graduation date connect with the registration form and the start date of the work experience certificate, and record the number of candidate items and scores |
| Work experience certificate | Whether start and end dates, position, seal, and handler's signature are complete and match the registration form dates | Adjust similarity to observe the key original text fragments retained after filtering, check the work experience certificate's start and end dates, position, employer seal, and handler's signature, and record the number of candidate items and scores |
| Exception record | Whether there is original record support and a reviewer's signature | Use different expressions of the same exception record to retrieve once each, compare candidates and scores, confirm that the original fragments of the registration form, education certificate, and work experience certificate can be traced, and then hand it to the responsible person for handling |
Three common errors
- Phenomenon: The name and graduation date of the same record do not match between the registration form and the education certificate. Cause: The versions of the two materials were not synchronized, and retrieval hit their respective fragments but did not perform field comparison.
- Phenomenon: The work experience certificate lacks the handler's signature or employer seal, and the review conclusion has no signature support. Cause: The material receipt was not archived, and the candidate fragments only covered the main text and not the signature/seal page.
- Phenomenon: The same question phrased differently cannot recall relevant materials, and although there are many candidates, the ones ranked at the front are all irrelevant. Cause:
searchModedoes not match the material form, and rerank was not enabled.
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 see the ranking of the top few; switch searchMode, adjust similarity and limit, record the number of candidate items and scores, record the number of citations separately by array length, and verify citation text tokens separately, without comparing them as equal to the number of items.
Business test: create an exception record, see whether the source materials can be traced, confirm that the fields, dates, and signatures of the registration form, education certificate, and work experience certificate are traceable, and after the reviewer signs, hand it to the responsible person for handling.
- In the knowledge base search test, switch
searchModefor the same exception record, record the number of candidate items and scores for semantic, full-text, and hybrid recall respectively, and check the fields, dates, and signatures of the registration form, education certificate, and work experience certificate. - Compare the top few rankings before and after enabling
usingReRank, adjustsimilarityandlimit, confirm that key original text fragments are still retained after trimming, record the number of citations by array length, and verify citation text tokens separately.