A Recommender System for Telecommunication Operators’ Campaigns

Abstract

Achieving a complete client characterization to attribute better and customized products is a trend on the rise. This information filtering process, known as “recommendation system”, increases companies revenues, and improve client quality-of-experience. Unfortunately, current state-of-the-art focus mostly on data sets with explicit client feedback regarding the advised product, or are more focused on case studies such as e-commerce, or movies recommendation. Our main goal is to understand the feasibility of a recommendation approach in one specific scenario: recommendation of telecommunication operators’ campaigns. We aim to determine the extent to which it is possible to characterize the clients, using implicit feedback and state-of-the-art recommendation algorithms.

We resorted to a data set supplied by an European Telecommunication Operator, with two years of advertised campaigns to a specific group of clients, and more than forty-six million lines of raw data. Having this data set, we applied pre-processing methods and analysed several algorithms to understand the feasibility of such approach in this real-world scenario. Results show that our approach can indeed infer the best product or service to advertise to the client – specially when we have historical data about past client adherence –, thus showing that these algorithms can improve recommendations in the context of a telecommunication operator.