What this category's data looks like
For multilingual materials and terminology handling in peer review organizations, the materials come from review invitations, expert review comments, and author revision notes, and the update cadence follows each round of submission for review and revision. When the three types of materials enter knowledge base retrieval and recall, their roles differ: review invitations provide the fields for the review target, language, and deadline; expert review comments carry terminology evaluations and revision suggestions; author revision notes record point-by-point responses, versions, and signed receipts. Multilingual materials often contain different translations of the same term, so after retrieval a unified terminology glossary and comparison draft need to be formed, determining that the same term is consistent throughout the text and that the translated name can refer back to the original text, before being handed over for external release. Field characteristics include name, quantity, date, signature, and receipt. When versions are not synchronized or signatures are missing, the same record can easily fail to match across two materials.
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 upper-limit trimming). The unified terminology glossary and comparison draft are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Review invitation | Whether the review target, language, deadline, and signed receipt are consistent with the same record in the expert review comments and author revision notes | In the knowledge base search test, switch searchMode, test once each with synonym-rewritten questions, record the number of candidates and scores, then check whether the name, quantity, date, and signature match |
| Expert review comments | Whether terminology evaluations and revision suggestions use a consistent translated name throughout the text, can refer back to the original text, and whether versions are synchronized | Turn usingReRank on and off to compare the ordering of the first few results, check whether candidates simultaneously contain semantic and full-text results, then sample three terms to check their usage throughout the text |
| Author revision notes | Whether point-by-point responses, versions, and signed receipts are complete, and whether the name, quantity, and date of the same record are consistent with the other two materials | Check whether the candidate snippets after similarity filtering retain the key paragraphs of the original text, separately record the number of items in the citation array, and check whether the tokens of the cited text exceed the budget |
| Unified terminology glossary and comparison draft | Whether the same term is consistent throughout the text, the translated name can refer back to the original text, and it can be handed over for external release | Sample three terms to check whether their usage is consistent throughout the text, and when signatures or receipts are missing, first check hits and associations |
Three common errors
- Phenomenon: The name, quantity, and date of the same record do not match between the review invitation and the author revision notes. Cause: Versions are not synchronized, the old draft and the new draft both enter the candidates, and checks are not made by hits and associations.
- Phenomenon: The translated names in the terminology glossary and comparison draft are inconsistent before and after, and cannot refer back to the original text. Cause: Multilingual materials have not unified terminology, retrieval candidates have not undergone secondary ranking, and the order of the first few results is unstable.
- Phenomenon: Expert review comments lack signatures or receipts, and this is discovered only before release. Cause: Material fields have not been completed, and upper-limit trimming of citations only affects the citation content budget and does not automatically fill in signatures.
How to verify the setup
Acceptance of this feature: test once each with synonym-rewritten questions, compare candidates and scores, then turn rerank on and off to inspect the ordering of the first few results; record the number of candidates and scores, and check the settings corresponding to the three identifiers searchMode, usingReRank, and similarity.
Business testing: sample three terms to check whether their usage is consistent throughout the text, confirm that the translated names can refer back to the original text, and check the name, quantity, date, and signature of the review invitation, expert review comments, and author revision notes.
- In the knowledge base search test, switch the retrieval mode, test once each with synonym-rewritten questions, record the number of candidates and scores, then turn rerank on and off to compare the order of the first few results.
- Sample three terms to check whether their usage is consistent throughout the text, check the name, quantity, date, signature, and receipt, and when versions are not synchronized, first check hits and associations before handing over for external release.