Jelenlegi hely

Intézeti szeminárium

Félév: 
2017/18 I. félév
Helyszín: 
Árpád tér 2. II. em. 220. sz.
Dátum: 
2017-10-10
Időpont: 
14:00-15:00
Előadó: 
Benczúr András (MTA SZTAKI, Budapest)
Cím: 
Recommender systems by traditional and online machine learning
Absztrakt: 
Recommender systems have to serve in online environments which can be 
highly non-stationary. Traditional recommender algorithms may periodically 
rebuild their models, but they cannot adjust to quick changes in trends 
caused by timely information. In our latest experiments, we observed 
that even a simple, but online trained recommender model can perform 
significantly better than its batch version. In the presentation, 
I will show online learning based recommender algorithms that can 
efficiently handle non-stationary data sets. I will discuss evaluation 
results over eight publicly available data sets. As part of our results, 
I will present our open source C++ recommender system with a scikit-learn 
style Python API, which is particularly suited for practical courses in 
recommender systems.