Rsdiv: Reranking for Multi-objective Optimized Recommender Systems

rsdiv provides the measurements and improvements for the multi-objective/diversifying tasks.

Some of its features include:

• various implementations of diversifying/ensemble reranking modules.
• various implementations of core recommender algorithms.
• evaluations for recommender systems from a quantitative/visual view.
• easy-to-use benchmarks for comparing and further analysis.
• automated hyperparameter optimization.

Installation

You can simply install the pre-build binaries with:

pip install rsdiv

More installation options can be found here.

Basic Usage

Prepare for a benchmark dataset

Evaluate the results in various aspects

Train and test a recommender

Reranking for diversity improvement

TODO

More diversifying algorithms

• implement the Bounded Greedy Selection Strategy, BGS diversify algorithm

• implement the Determinantal Point Process, DPP diversify algorithm

Hyperparameter optimization

• compatible with Optuna.

Ensemble ranking

• support the ensemble ranking modules

For developers

Contributions welcome! Please contact us.

During your development stage, make sure you have pre-commit installed in your local environment:

pip install pre-commit
pre-commit install