FORTE: A Production Framework for Feature Evaluation and Deprecation in Industrial Recommender Systems
arXiv cs.AI · 10 Oct 2026, 02:18 AEST
FORTE is a model-agnostic production framework designed for feature evaluation and deprecation in recommender systems. It enables efficient daily feature evaluation and identifies deprecable feature sets without causing online regressions.
Key facts
- Production recommender systems have feature-lifecycle asymmetry: adding new features must produce gains on some metric, while feature removal must not regress any metric.
- We present FORTE, a model-agnostic production framework for discovering, validating, and removing features.
- We validate FORTE over 2 large-scale recommender systems and find that it identifies deprecable feature sets which do not cause online regressions, while the impact on efficiency depends on model characteristics.
- On a search-relevance model, FORTE resulted in 3 nested candidate sets of 32, 48, and 59 features.
- Its launch reduced the trainable parameters by 29.9%, improved serving throughput by 19.8-25.5%, and reduced annual training costs by hundreds of thousands of dollars.
Read aloud
FORTE is a model-agnostic production framework designed for feature evaluation and deprecation in recommender systems. It enables efficient daily feature evaluation and identifies deprecable feature sets without causing online regressions. Production recommender systems have feature-lifecycle asymmetry: adding new features must produce gains on some metric, while feature removal must not regress any metric. We present FORTE, a model-agnostic production framework for discovering, validating, and removing features. We validate FORTE over two large-scale recommender systems and find that it identifies deprecable feature sets which do not cause online regressions, while the impact on efficiency depends on model characteristics. On a search-relevance model, FORTE resulted in three nested candidate sets of thirty two, forty eight, and fifty nine features. Its launch reduced the trainable parameters by twenty nine point nine percent, improved serving throughput by nineteen point eight-twenty five point five percent, and reduced annual training costs by hundreds of thousands of dollars.
Sources in this cluster
- FORTE: A Production Framework for Feature Evaluation and Deprecation in Industrial Recommender Systems
arXiv cs.AI · 10 Oct 2026, 02:18 AEST