What this category's data looks like
This page addresses three types of materials in graduate training process management: training plans, proposal reports, and mid-term assessment forms. The sources are formal documents and process records submitted at each training stage, with an update cadence that rolls forward with semester progress and assessment milestones. Field characteristics include name, student ID, course name, credit count, proposal date, mid-term assessment date, supervisor signature, assessment panel signature, and receipt status. When materials enter the question acceptance and missing-item follow-up function, the training plan serves as the baseline for courses and credits, the proposal report serves as the basis for topic selection and time milestones, and the mid-term assessment form serves as the basis for progress and signature receipts. Knowledge base retrieval and recall is responsible for retrieving candidate snippets from these materials when accepting a question, so that subsequent nodes can determine missing items and generate follow-up questions to fill them in.
Configuration and implementation checks
This stage 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). Acceptance records and the follow-up questions to fill gaps are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Training plan | Whether the course name, credit count, study date, and supervisor signature correspond consistently with the proposal report and mid-term assessment form | In the search test, use the course name and credits as the query to retrieve, verify the name, count, date, and signature in the candidate snippets, record the number of candidates and scores, and switch searchMode to compare semantic, full-text, and hybrid results |
| Proposal report | Whether the topic title, proposal date, assessment panel signature, and receipt match those in the training plan and mid-term assessment form | Use the topic and date as the query to retrieve, verify the name, date, and signature in the candidate snippets, check whether multi-path merging includes both semantic and full-text candidates, and record the number of candidates and scores |
| Mid-term assessment form | Whether the assessment date, progress description, supervisor signature, and receipt are synchronized with the proposal report and training plan version | Use the assessment date and progress as the query to retrieve, verify the date, signature, and receipt in the candidate snippets, enable rerank to compare the order of the top few results, and record the number of candidates and scores |
| Retrieval node | Whether the candidate snippets cover the fields required for missing-item determination, and whether key original snippets are retained after similarity filtering and citation limit trimming | Test once each with synonymous rewritten queries, compare candidates and scores, toggle rerank to observe the ordering of the top few results, separately record the array count and citation text tokens, and verify the effects of similarity, limit, and usingReRank |
Three common errors
- Symptom: The same course name does not match between the training plan and the proposal report. Cause: The two material versions are not synchronized, and retrieval candidates hit only one of them.
- Symptom: The mid-term assessment form lacks a supervisor signature or receipt, yet the question is answered directly. Cause: The missing item was not identified, and follow-up to fill it in was not performed first.
- Symptom: There are many candidates, but the ones ranked at the top are irrelevant, and the answer lacks key passages. Cause: Rerank was not enabled or similarity filtering was too loose, and citation limit trimming did not retain key original snippets.
How to verify the setup
Acceptance for this function: Test once each with synonymous rewritten queries, compare candidates and scores, then toggle rerank to observe the ordering of the top few results; record the number of candidates and scores, separately record the array count and citation text tokens, and verify the effects of similarity, limit, searchMode, and usingReRank.
Business test: Ask a question missing key information and see whether it follows up first before answering.
- In the knowledge base search test, switch
searchModeto compare the candidate results and scores of the semantic, full-text, and hybrid modes. - Compare the ordering of the top few results before and after enabling rerank, and verify whether key original snippets are retained after similarity filtering and citation limit trimming.