# Speaker
Katharina Strecker
Center for Solar Energy and Hydrogen Research
Baden-Württemberg (ZSW)
# Time and location
Wednesday, December 10, 16:00
1101.B417 (Welfengarten 1)
# Title
Leveraging Recommender Systems for Automated Feature Engineering
#Abstract
A persistent gap in today’s AutoML systems is their limited support for automated feature engineering (AutoFE) particularly in the data selection phase, deciding which external data sources should be integrated as new columns. This step is crucial for tabular ML performance, yet it remains largely manual, heavily domain-dependent, and difficult to automate with existing AutoML optimization techniques. This talk introduces a new approach that reframes data selection as a recommender-system (RecSys) task: datasets act as users, feature categories as items, and feature importance as feedback. Using topic modeling and a LightGCN model, our AutoFE RecSys recommends relevant new features directly from raw tables. Evaluated on a new benchmark of 280 energy datasets, it performs on par with LLM-based baselines. This work highlights how RecSys methods can extend AutoML upstream into data acquisition and feature discovery.
