1Cademy - MiniLM Encoder + CJK-Only Queries on MOOC-CS: Hierarchical R@10 Rises from 23.1 to 26.5 with Flat Dense at 21.7

Learn Before

MOOC-CS: Language-Matched Controls (Results) in Auditable Strict-Parity Evaluation of Prerequisite-Graph Retrieval for RAG under Leakage Controls
Method Part 1: Multilingual Encoder + CJK Query Rewrite as a MOOC-CS Control (Auditable Strict-Parity Graph-RAG Paper)

Example

MiniLM Encoder + CJK-Only Queries on MOOC-CS: Hierarchical R@10 Rises from 23.1 to 26.5 with Flat Dense at 21.7

On MOOC-CS ( $n = 103$ ) with the English MiniLM dense encoder held fixed, switching the query side from English templates to CJK-only concept-name queries raises flat-dense Recall@ $10$ from 16.0 [11.5, 20.9] to 21.7 [17.0, 27.0], the hierarchical baseline from 23.1 [17.2, 29.3] to 26.5 [20.1, 33.7], and Adaptive (heuristic) from 23.1 to 26.4. This row isolates the effect of query cleaning alone under an unchanged English-only encoder.

0

1

Updated 2026-05-17

Contributors are:

Who are from:

References

Reference: Auditable Strict-Parity Evaluation of Prerequisite-Graph Retrieval for RAG under Leakage Controls

Learn Before

Related