Contextual Bandit Models for Long- and Short-Term Recommendations

Recommender systems aim to capture interests of users to provide tailored recommendations. User interests are often unique and depend on many unobservable factors including internal moods or external events. We present a unified contextual bandit framework for recommendation problems that is able to capture both short- and long-term interests of users. The model is devised in dual space and the derivation is consequentially carried out using Fenchel-Legrende conjugates and thus leverages to a wide range of tasks and settings. We detail two instantiations for regression and classification scenarios and obtain well-known algorithms for these special cases. The resulting general and unified framework allows for quickly adapting contextual bandits to different applications at-hand. The empirical study demonstrates that the proposed short- and long-term framework outperforms both, short-term and long-term models. Moreover, a tweak of the combined model proves beneficial in cold start problems.

Bio

Maryam Tavakol is in the last year of her PhD studies in Machine Learning at TU Darmstadt under joint supervision of prof. Ulf Brefeld and Jo­hannes Fürnkranz, whilst working as a research assistant in Machine Learning group at Leuphana University of Lüneburg. The main area of her research during PhD is to use machine learning techniques, particularly, Reinforcement Learning in sequential recommendation systems which has led to novel contributions in the area of recommendation. Before that, she received both of her bachelor and master degrees in Computer Science from the university of Tehran in Iran.
She also has a 6-month internship experience in recommender system group of Criteo company in Paris.