What this category's data looks like
In the scenario of regular higher education institution admissions enrollment and filing, the materials entering the knowledge base retrieval and recall stage mainly include three categories: enrollment plans, filing rules, and admission rosters. Enrollment plans carry fields such as institution-major groups, planned numbers, selected examination subjects, and batches, and updates are synchronized with plan adjustment notices; filing rules carry fields such as filing ratios, tie-breaking sorting, adjustment and withdrawal conditions, and updates are synchronized with annual policy documents and supplementary explanations; admission rosters carry fields such as candidate numbers, names, admitted majors, admission statuses, and dates, and updates advance daily with the admission process of each batch. The three categories of materials serve as sources of candidate fragments in the retrieval node, and according to material form they respectively go through semantic, full-text, or hybrid recall, and after merging and reranking, cited content is obtained for subsequent nodes or manual drafting of issuable materials and handover records to retrieve.
Configuration and implementation checks
This stage is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate fragments and similarity scores (after pattern, merging, reranking, filtering, and upper-limit trimming). The issuable material drafts and handover records are then completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Enrollment plan | Fields complete: institution-major groups, planned numbers, selected examination subjects, batches; dates and signatures consistent with the plan adjustment notice | In the knowledge base search test, switch searchMode to semantic, full-text, or hybrid, check whether the candidate fragments cover the above fields, and record the similarity scores |
| Filing rules | Fields complete: filing ratios, tie-breaking sorting, adjustment and withdrawal conditions; version and dates consistent with the annual policy documents | Compare the order of the top few items before and after enabling usingReRank, check whether the rule clauses fall in the front positions, and inspect the version identifier and release date |
| Admission roster | Fields complete: candidate numbers, names, admitted majors, admission statuses, and dates; signatures and receipts complete | Search by candidate number or major name, check whether the name, quantity, and date of the same record in the roster and the handover record are consistent, and whether missing items are blocked |
| Handover record | Responsible person clear, version and dates consistent, receipt can be signed for | Check the number of candidate citations and the returned results under the limit and similarity settings, and confirm that the tokens of the cited text and the number of array items are recorded separately |
Three common errors
- Phenomenon: The major name, planned number, or date for the same candidate in the admission roster and the handover record do not match. Cause: The versions of the two materials are not synchronized, and the retrieved hit fragments come from different batches or different revisions.
- Phenomenon: The handover record lacks a signature or receipt, and the next stage cannot sign for it. Cause: The material draft fields are incomplete, the responsible person column is empty, and the retrieval candidates do not cover the signature column.
- Phenomenon: The same question cannot recall the rule clauses when phrased differently. Cause: The retrieval mode does not match the material form, question optimization does not cover references and expansions, and candidates are trimmed by similarity filtering.
How to verify the setup
Acceptance of this function: test once each with synonymously rewritten questions, and compare the number of candidates and similarity scores; then toggle reranking on and off to see whether the order of the top few items changes.
Business test: draft one according to the template and complete one handover, and see whether missing items are blocked.
- In the knowledge base search test, switch
searchMode, compare the number of candidates and scores under the three modes of semantic, full-text, and hybrid, and confirm that the material form matches the mode. - Adjust
similarityandlimit, enable or disableusingReRank, record the tokens of the cited text and the number of array items, and check whether key fragments are still retained.