DataRec Library for Reproducible in Recommend Systems

The rapid growth of recommender‑system research has outpaced the tools needed to keep experiments reproducible. Researchers often juggle dozens of public benchmarks—MovieLens, Amazon reviews, Last.fm, Gowalla—while manually handling downloads, format conversions, and version control. Small changes in preprocessing or dataset version can swing accuracy scores, making it difficult to compare new algorithms fairly. DataRec, an open‑source Python library, was created to close this gap by providing a single, standardized interface for acquiring, verifying, and preparing recommendation datasets, turning a fragmented workflow into a reliable, repeatable process.

DataRec automates the entire lifecycle of a dataset. A single command downloads the original file, checks its SHA‑256 checksum against a stored reference, and flags any upstream modifications. The library ships with ready‑made readers for CSV, TSV, and JSON formats, exposing every benchmark as a uniform DataRec object. Built‑in filters let users retain users with a minimum interaction count, discard rare items, and apply temporal splits that mimic real‑world training‑test scenarios. Because the API is dataset‑agnostic, the same preprocessing pipeline can be reused across MovieLens, Amazon, or custom private collections without rewriting code.

By enforcing consistent data provenance, DataRec dramatically reduces the time researchers spend on boilerplate tasks and lowers the risk of hidden bugs that skew results. Compared with heavyweight reproducibility frameworks that impose full pipelines, DataRec offers a lightweight, plug‑and‑play approach that integrates seamlessly into existing codebases, making rapid prototyping and incremental experiments straightforward. The library’s open‑source model encourages community contributions of new datasets and splitting strategies, ensuring it stays current as the field embraces large language models and graph‑based recommenders. Ultimately, DataRec empowers both academia and industry to benchmark innovations on a common, trustworthy foundation.