Authors:
Kiana Liaghat (Lead Machine Learning Engineer, Jabama, Tehran, Iran)
Mohammad Forouhesh (Data Science Manager, Jabama, Tehran, Iran)
Contact:
This repository contains the preprint draft of our ongoing research on multi-metric optimization in ranking systems. Optimizing for a single metric often fails to capture the complexity of real-world business goals and user needs. By optimizing across multiple—sometimes competing—business metrics, our approach offers a more accurate modeling of the environment, leading to better recommendation and ranking outcomes for both users and businesses.
Our work introduces a mathematically rigorous and computationally efficient algorithm for multi-objective optimization. It supports:
- Arbitrary numbers of business and user-centric KPIs.
- Ability to incorporate explicit stakeholder-defined importance weights.
- Theoretical optimality using generalized eigenvalue problems and Rayleigh quotient formulation.
🚧 This is a preprint draft.
We are actively:
- Finalizing the writing and structure of the paper.
- Completing experimental validation and A/B testing.
- Refining implementation details and performance evaluation.
Please note that this version may undergo significant revisions as the work progresses.
Multi_Metric_Optimization_PrePrint.pdf: The draft preprint of our paper.- Future updates will include:
- Experimental results
- Final camera-ready version
If you find this work useful or would like to reference it, please use the following citation (placeholder until official publication):
@article{liaghat2025multimetric, title={Multi-Metric Optimization: Fast and Flexible Re-ranking Over Multiple Business Metrics}, author={Liaghat, Kiana and Forouhesh, Mohammad}, journal={arXiv preprint}, year={2025} }
Ranking · Multi-objective Optimization · Recommendation Systems · Rayleigh Quotient · Generalized Eigenvalue Problem · Business KPIs
We welcome questions, feedback, and collaboration. Feel free to contact us at the emails listed above.