Ctx-effSAMWMIX: A Contextual Multi-Armed Bandit Algorithm for personalized recommendations

Abstract

Machine Learning algorithms play an active role in modern day business activities and have been put to an extensive use in the marketing domain as well. In Ecommerce domain, these algorithms play an important role in suggesting recommendations to users, be it a merchandise of interest to the user or a news article for a website visitor. Due to the larger variety of available information and multiplicity in the merchandise based data, these personalized recommendations play a major role in the successful business activity that could be a sale in the case of an Ecommerce website or a click on a news article in case of a news website. The personalized recommendation problem, where the challenge is to choose from a set of available choices to cater to a target user group, can be modelled as a Contextual Multi-Armed Bandit problem. In this work we propose Ctx-effSAMWMIX which is based on LinUCB and effSAMWMIX algorithms. We empirically test the proposed algorithm on Yahoo! Frontpage R6B dataset by using an unbiased offline evaluation technique proposed in literature. The performance is measured on Click Through Rate (CTR) which effectively reports the ratio of Clicks the recommended articles obtained to that of total recommendations. We compare the performance of Ctx-effSAMWMIX with LinUCB and a random selection algorithm and also report the results of t-tests performed on the mean CTRs.