We released OmniRec 1.0.0, the first major version of our open-source Python library for reproducible and interoperable recommender systems experimentation.
OmniRec was recently presented by Philipp Meister at ECIR 2026 in Delft, where he introduced the library and our accompanying demo paper. Since then, we have continued working on the library and have now moved from version 0.2.0 to version 1.0.0. While this release primarily focuses on bug fixes and stability improvements, it also introduces several new features.
Lukas Wegmeth, Moritz Baumgart, Philipp Meister, Bela Gipp, and Joeran Beel. 2026. OmniRec: The All-In-One Solution for Reproducible and Interoperable Recommender Systems Experimentation. In Advances in Information Retrieval: 48th European Conference on Information Retrieval, ECIR 2026, Delft, The Netherlands, March 29 – April 2, 2026, Proceedings, Part IV. Springer-Verlag, Berlin, Heidelberg, 129–135. https://doi.org/10.1007/978-3-032-21321-1_18
OmniRec
For those not being familiar with OmniRec, here a short introduction.
OmniRec is unique mainly because it is not just another recommender-system toolkit; it is a unifying layer for the whole experimental workflow. It addresses a concrete reproducibility problem in recommender-systems research: fragmented data handling, inconsistent preprocessing, weak interoperability, and poor dataset referencing. Its core contribution is to standardize access to more than 230 datasets, canonicalize data into a common structure, and let researchers define preprocessing pipelines once rather than repeatedly adapting them for each framework or experiment.
What makes OmniRec especially distinctive is the combination of dataset standardization, cross-framework execution, and reproducible evaluation. It integrates multiple established recommender libraries, including RecPack, RecBole, LensKit, and Elliot, while using unified runner interfaces and isolated execution environments to reduce dependency conflicts. It also centralizes evaluation metrics, checkpointing, caching, experiment resumption, and result storage, so comparability is built into the workflow rather than left to the researcher’s ad hoc scripts. In blunt terms: OmniRec’s uniqueness is not a new recommendation algorithm; it is an experiment infrastructure that makes existing algorithms and datasets more comparable, reusable, and reproducible.
More Flexible Hyperparameter Search
OmniRec 1.0.0 introduces new Plan Components for defining hyperparameter search strategies in experiment plans.
In addition to fixed parameter values, users can now explicitly define parameters for grid search or random search. This makes it easier to run systematic experiments over multiple configurations without manually defining each run. For large search spaces, the new random search components allow users to sample from selected values or ranges rather than evaluating every possible combination.
This improves the flexibility of OmniRec’s experiment planning while keeping configurations concise and readable.
Easier Identification of Experiment Runs
Another important improvement concerns the traceability of experiment results.
Previously, individual runs were identified by hash values, but it was not always straightforward to see which hyperparameter configuration belonged to which result. OmniRec now explicitly stores the algorithm configuration along with the checkpoints, making it much easier to inspect completed runs and map results back to their original settings.
The configuration hash now also includes the random seed, preventing collisions between runs that use the same dataset and algorithm settings but differ in their random state.
Lineage Tracing for Preprocessing Steps
The largest new feature in this release is lineage tracing for preprocessing steps.
OmniRec now records detailed information about how a dataset was processed, including the preprocessing components used, their parameters, execution metadata, system information, and dataset shape before and after each step. This makes it easier to understand where a dataset came from, how it was transformed, and under which conditions it was created.
This information can be stored alongside the dataset in OmniRec’s RSDS file format, which is now better versioned to support backwards compatibility and future extensions. To make this metadata easier to inspect, RecSysDataSet also provides new formatting methods for dataset details and lineage information.
Further Improvements
In addition to these features, OmniRec 1.0.0 includes a broad range of fixes and usability improvements. Among others, the release improves result extraction through a new evaluator method, preserves timestamps during implicit feedback conversion, improves coordinator-runner communication, and resolves several issues related to managed environments and runner stability. Click here for a full changelog.
The documentation has also been extended with new examples for provenance handling, result extraction, custom algorithms, plan components, and related APIs.
Outlook
With version 1.0.0, OmniRec takes an important step toward more robust and reproducible recommender systems research. The release strengthens experiment tracking, improves dataset provenance, and makes hyperparameter exploration more flexible, while also addressing many stability issues reported since the previous release.
We invite researchers, students, and practitioners to try OmniRec 1.0.0, contribute to its further development, and send us feedback on their experiences with the library.
0 Comments