Contextual bandit algorithm called LinUCB / Linear Upper Confidence Bounds as proposed by Li, Langford and Schapire