What this category's data looks like
This category's data typically consists of three types of materials: training notices, credit recognition forms, and participant lists, often containing multilingual translated names and specialized terminology. Training notices provide the project name, credit hours, and assessment requirements; credit recognition forms record the recognized subject, credit value, and recognition date; participant lists state teacher names, units, and sign-in status. Materials are updated by training batch; within the same batch, notices are issued first, followed by the list and recognition form in sync, with frequent version iterations. In terms of field characteristics, the same term may appear in different materials as an abbreviation, full name, or translated name, and date formats and signature fields are also inconsistent. When entering knowledge base retrieval and recall, training notices, credit recognition forms, and participant lists each participate as independent sources in candidate recall, providing original text snippets and similarity evidence for the unified terminology table and comparison draft.
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 pattern, merge, rerank, filter, and limit trimming). The "unified terminology table and comparison draft" is completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Training notice | The project name, credit hours, and assessment requirements are internally consistent within the notice and can cross-reference the recognition form and list | In the search test, switch searchMode to semantic, full-text, and hybrid once each, and check whether the project name and credit hour wording in the notice candidate snippets are consistent |
| Credit recognition form | The recognized subject, credit value, and recognition date are consistent within the form, signature and receipt fields are complete, and they match the notice and list | Use synonymous questions for the same term to retrieve once each, compare the number of candidates and scores, and check whether the subject names and dates in the recognition form match the list |
| Participant list | Teacher names, units, and sign-in status are consistent within the list, and dates and signatures can cross-reference the notice and recognition form | Run once each with usingReRank enabled and disabled, observe changes in the ranking of the top list candidates, and check whether the number of people and dates in the list and recognition form match |
| Unified terminology table and comparison draft | The same term is written consistently throughout, translated names can cross-reference the original text, and statistical tables or authorization results are not mixed in | In the retrieval results, observe the number of candidates and scores by similarity and limit, and extract three terms to check whether their spelling is consistent throughout |
Three common errors
- Phenomenon: The same teacher's name and unit do not match between the participant list and the credit recognition form. Cause: The two materials were imported separately and versions were not synchronized, so retrieval hits snippets from different batches.
- Phenomenon: The project name in the training notice does not match the subject name in the recognition form. Cause: Abbreviations and full names of terms are mixed, and candidate recall does not normalize by the same term.
- Phenomenon: The recognition form lacks signatures or receipts, and the list lacks sign-in dates. Cause: Materials entered the knowledge base before the finalization process was complete, and the snippets returned by retrieval themselves have incomplete fields.
How to verify the setup
Acceptance for this feature: test once each with synonymously rewritten questions, compare candidates and scores, then toggle rerank on and off to see the ranking of the top few.
Business test: extract three terms and check whether their spelling is consistent throughout.
- In the knowledge base search test, switch between semantic, full-text, and hybrid modes respectively, record the number of candidates and scores, and toggle rerank on and off to compare the order of the top few.
- Check whether the names, quantities, dates, and signature receipts of the same record match across the training notice, credit recognition form, and participant list.