Knowledge Base Retrieval for District Teaching Research and Classroom Reviews: Trace Exceptions and Review Results

Teaching Reviews: review retrieval for exception review. Compare the source records, required inputs and downstream results before handing the task over.

What this category's data looks like

District and county teaching research activity and classroom observation and evaluation materials usually consist of three types: teaching research activity notices, classroom observation records, and classroom evaluation minutes. Notices contain the activity name, time, location, participating schools, and teacher roster; classroom observation records contain the instructor, class, topic, observation points, and time; classroom evaluation minutes contain review opinions, conclusions, and signature fields. The three types of materials are updated by activity batch, and the same activity may appear repeatedly in notices, records, and minutes. When entering the knowledge base retrieval and recall function, notices provide the activity background and scope, classroom observation records provide the classroom process and observation points, and classroom evaluation minutes provide conclusions and responsible signatures. The fields, dates, and signatures of each material need to be mutually corroborating. During retrieval, the recall method is selected according to the material form so that candidate snippets cover multiple sides of the same activity, providing clues that can be traced back to the source materials for anomaly tracing and manual review.

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 mode, merge, rerank, filter, and limit trimming). The anomaly list and review conclusions are completed by subsequent nodes or manually.

ObjectCheckVerification action
Teaching research activity noticeFields include activity name, date, participating schools, and teacher roster; signatures and receipts are completeIn the search test, switch searchMode to hybrid and check whether candidates include both semantic and full-text results
Classroom observation recordInstructor, class, topic, observation points, and date can correspond to the notice rosterSearch once each using different phrasings of the same record, and compare the number of candidates and the similarity scores
Classroom evaluation minutesReview opinions, conclusions, and signature fields are complete, and the date is consistent with the classroom observation recordCompare the ranking of the top few items before and after enabling usingReRank, and record the number of citations after limit trimming
Knowledge base retrieval serviceOutputs candidate snippets and similarity scores, after mode, merge, rerank, filter, and limit trimmingCheck whether candidates include both semantic and full-text results, and record cited text tokens and array item counts separately

Three common errors

  • Phenomenon: The same activity name is written differently in the notice and minutes, and the dates differ by several days. Cause: The two materials were entered separately and were not aligned by activity number.
  • Phenomenon: The classroom observation record has been updated, but the classroom evaluation minutes are still the old version, and the conclusion cites old observation points. Cause: The version was not synchronized, and the reviewer was not notified of the update.
  • Phenomenon: The classroom evaluation minutes lack a signature or receipt, and the responsible person for the anomaly record cannot be confirmed. Cause: The signature field was not filled in, and the receipt was not archived.

How to verify the setup

Acceptance for this function: test once each with synonymously rewritten questions, compare candidates and scores, then toggle rerank on and off to inspect the ranking of the top few items.

Business test: create an anomaly record and see whether it can be traced back to the source materials.

  • In the search test, switch searchMode and compare the number of candidates and scores among semantic, full-text, and hybrid.
  • Enable and disable usingReRank, observe changes in the ranking of the top few items, and check whether the cited text after limit trimming retains key snippets.