AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

On the Generalizability and Predictability of Recommender Systems

Duncan C. McElfresh, Sujay Khandagale, Jonathan Valverde, John P Dickerson, Colin White

NeurIPS 2022 · Learning theory and algorithms › Prediction and decision losses · approach: Ranking models from comparisons · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use meta-feature-based algorithm selection instead of prior algorithm-selection approachesRecZilla versus cf4cf-meta and cunha2018 · 19 test sets, PREC@10%Diff: 52.9$\pm 23.0$ -> 33.2$\pm$ 22.8; 43.5 \pm 21.8 -> 33.2$\pm$ 22.8Improved
Use meta-feature-based algorithm selection instead of prior algorithm-selection approachesRecZilla versus cf4cf-meta and cunha2018 · 19 test sets, PREC@10PREC@10 of best prediction: 0.00808\pm 0.00773 -> 0.00915$\pm$ 0.00840; 0.00813\pm 0.0113 -> 0.00915$\pm$ 0.00840Improved
Add dataset families to meta-trainingMeta-learners predicting PREC@10 · Leave-one-dataset-family-out evaluationMAEImproved
Transfer hyperparameters tuned on one dataset to another datasetRecommender algorithms · Cross-dataset hyperparameter transfernormalized performanceHurt
Choose algorithms and hyperparameters separately for each dataset and metric rather than use one universal choiceRecommender systems · 85 datasets and 315 metricsalgorithm rankImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy