**ColBERTv2/RePlug reranking** is included as a non-graph reranking baseline in this paper's strict-parity baseline set for prerequisite retrieval. The dense seed pool produced by the shared query encoder is reranked using a **ColBERTv2** late-interaction scorer and/or a **RePlug**-style retrieval-augmented scorer over the same candidate concepts. Under the strict-parity contract, this baseline shares the same encoder, $$m/k$$ budgets, split, and exact-ID matching rule as the other compared systems, so any reported difference is attributable to the reranking step rather than to an interface change. It serves as the comparison point for whether neural reranking — without graph traversal — closes the gap to hierarchical prerequisite retrievers.

ColBERTv2/RePlug Reranking Baseline for Strict-Parity Prerequisite Retrieval

**Truncated personalized PageRank (truncated PPR)** is included as a graph-aware comparator in this paper's strict-parity baseline set. Starting from a seed distribution over the dense top-$$m$$ candidates, personalized PageRank scores are computed on the prerequisite graph and the propagation is **truncated** (e.g., after a bounded number of power-iteration steps or below a small score threshold) so that the score concentrates near the seeds rather than spreading across the whole graph. Under the strict-parity contract, the baseline shares the same encoder, $$m/k$$ budgets, split, and exact-ID matching rule as the other compared systems, so it isolates truncated-PPR diffusion as the graph policy. It sits alongside the LightRAG-style baseline as one of the two best-performing comparators on MOOC-CS under the default English MiniLM configuration.

Truncated Personalized PageRank Baseline for Prerequisite Retrieval

The **matched-budget best-first traversal comparator** is a graph-aware strict-parity baseline in which the prerequisite graph is traversed by expanding the highest-scoring frontier node first, subject to the same node/edge/token budget as the paper's diffusion retriever. Under the strict-parity contract, it shares the same encoder, $$m/k$$ budgets, split, and exact-ID matching rule as every other compared system, so any difference between this comparator and the paper's diffusion policy is attributable to *the graph policy itself* (best-first traversal versus dense-seeded local diffusion with role-aware quotas) rather than to budget or interface differences.

Matched-Budget Best-First Traversal Comparator in Strict-Parity Prerequisite Retrieval

On **MOOC-CS**, the paper additionally runs a configuration that uses a **multilingual encoder** plus a **CJK-only query rewrite** to handle Chinese-character queries. This configuration is treated as a **control experiment**, not as a headline strict-parity comparator: because it changes the encoder (and therefore the parity contract), its numbers are reported separately and are used only to diagnose how much of the MOOC-CS gap is attributable to language alignment rather than to graph policy.

Multilingual Encoder + CJK Query Rewrite as a MOOC-CS Control

This organizer keeps part 2 of the method details for Baselines small enough for learners to study without a long flat child list.

Google

This organizer groups method, retrieval, ranking, and graph-structure details used by Baselines.

Method Details for Baselines (Auditable Strict-Parity Graph-RAG Paper)

Root URL: https://1iman.com/
Canonical paper URL: https://1iman.com/assets/working-papers/pdf/emnlp-auditable-strict-parity-evaluation-of-prerequisite-graph-retrieval-for-rag-under-leakage-controls.pdf
Portfolio PDF URL: https://1iman.com/assets/working-papers/pdf/emnlp-auditable-strict-parity-evaluation-of-prerequisite-graph-retrieval-for-rag-under-leakage-controls.pdf
Paper title: Auditable Strict-Parity Evaluation of Prerequisite-Graph Retrieval for RAG under Leakage Controls
Use this Reference node when citing paper-derived 1Cademy proposals from the Researcher agent.

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

This organizer keeps part 1 of the method details for Baselines small enough for learners to study without a long flat child list.

Learn Before

Related

Learn After