Knowledge Base Retrieval for Undergraduate Curriculum Planning: Coordinate Operations and Application Handover

Program Revision: review retrieval for operational handover. Check source versions, input mappings and the values received by downstream nodes.

What this category's data looks like

The materials for undergraduate program revision include the program plan, course syllabi, and the credit structure table. Sources are academic affairs system exports and departmental submissions, with an update cadence by revision batch, and fields include course name, credits, class hours, semester offered, signature, and date. After the program plan enters the knowledge base, it serves as the primary retrieval object, the course syllabi serve as supporting fragments, and the credit structure table serves as the reference basis. In retrieval, the three types of materials respectively assume the roles of primary document, supporting document, and verification document; candidate fragments and similarity scores are output after mode, merge, re-rank, filter, and upper-limit trimming, and the business deliverable is generated by subsequent nodes or manually based on this.

Configuration and implementation checks

This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate fragments and similarity scores (after mode, merge, re-rank, filter, and upper-limit trimming). The handover-ready deployment status and operation records are then completed by subsequent nodes or manually.

ObjectCheckVerification action
Program planFields, dates, and signatures are complete; course names match the credit structure table item by itemIn the search test, use searchMode to switch between semantic, full-text, and hybrid, check candidate fragments and scores, and record the number of candidates
Course syllabiCourse names, credits, and class hours match the corresponding records in the program plan, and versions are in syncTurn usingReRank on and off to compare the ordering of the top few results, and check whether the syllabus and program plan fields correspond
Credit structure tableCredit totals match the program plan and course syllabi item by item, and dates and signatures are not missingCheck signatures and receipts against the handover form, and record candidate scores under the similarity threshold and the tokens of the cited text
Deployment status and operation recordsVersion, connection, and task records are consistent across all threeCheck the version number, connection status, and task records item by item, and fill in whichever record is missing

Three common errors

  • Symptom: The name, credits, and date of the same course do not match between the program plan and the course syllabus. Cause: The two materials were revised separately, versions were not synchronized, and signatures or receipts are missing.
  • Symptom: The credit structure table totals are inconsistent with the program plan, and irrelevant candidate fragments rank at the top. Cause: The retrieval mode does not match the material form, and re-ranking is not enabled.
  • Symptom: The answer lacks key paragraphs, and the number of citations is small but the content is incomplete. Cause: Similarity filtering and citation upper-limit trimming are too tight, and the tokens of the cited text were not checked.

How to verify the setup

This feature acceptance: test once each with synonym-rewritten questions, compare candidates and scores, then turn re-ranking on and off to observe the ordering of the top few results; switch searchMode and usingReRank, and record the number of candidates and the similarity scores.

Business testing: check once item by item against the handover form to see which record is missing; check the fields, dates, and signatures of the program plan, course syllabi, and credit structure table, and confirm that version, connection, and task records are consistent across all three.

  • In the knowledge base search test, switch retrieval modes, compare candidate results and scores, and record the number of candidates.
  • Turn re-ranking on and off, compare the order of the top few results, and check the tokens of the cited text and the number of array items.