Introduction of temporal and spatial dimensions into a recommender system

Abstract

Dieses Papier befasst sich mit einer Vorhersagemodellierung Problem und stellt die Werkzeuge und die Vorgehensweise, um es zu bekämpfen.

This paper deals with a predictive modelling problem and presents the tools and the approach to tackle it. More precisely, it starts out from a recommender engine operating in supermarkets, and especially from the model predicting whether the customers will use the discount they got from a coupon. This project aims to introduce temporal and spatial dimensions into this model, since customers may have different behaviour according to these two aspects. Improving a predictive model first means to set the proper indicators in order to evaluate its performance. Different metrics are thus introduced on that purpose and selected mainly regarding their suitability towards our business problem. Furthermore, the main highlight of this thesis is made on comparing different models by assessing their predictive power. The temporal and spatial dimensions are successively introduced, by modifying the inputs of the current model and keeping the same method. Through the performance indicators previously defined, we assess whether a new dimension is worth being kept, while getting some insights about the customers' behaviour.