Replication package for "RAG-Reviewer: A Retrieval-Augmented Generation Framework for Automated Code Review Comment Generation"

This repository contains artifacts for the experiments described in RQ1–RQ3.<b>fine_tuned_checkpoints.zip</b><br>Checkpoints for each experimental run. Each folder follows the naming convention:<br><code>modelName_ragStrategy_finetuned_best_ckp_epoch</code>, where:<code>modelName</code> refers to the backbone model (e.g., CodeT5, CodeReviewer).<code>ragStrategy</code> denotes the prompting strategy: <code>rag_pair</code>, <code>rag_singleton</code>, or <code>vanilla</code>.<code>epoch</code> indicates the best-performing epoch based on validation performance.<b>code_embeddings.zip</b><br>Vector embeddings of source code from the Tufano et al. dataset (train/val/test splits).<br>Embeddings are stored as <code>.pkl</code> files and used for retrieval tasks.<b>rag_candidate.zip</b><br>Contains the top-K most similar training examples for each instance in the train, validation, and test sets.<br>These candidates were selected using semantic similarity for retrieval-augmented generation (RAG).The source code for reproducing these experiments is available on GitHub:<br>https://github.com/RAG-Reviewer/RAG-Reviewer