Building a recommender system for a B2B company - Etilux Case

Abstract

Nowadays, data analysis becomes crucial inside companies. They usually detain a lot of data but do not necessarily use or analyse them. Business analytics tools have been developed over the years to help companies making strategic decisions. These tools are helping firms increasing their revenues or target more efficiently their customers. It enables the discovery and usage of new kind of knowledge. Among those tools, recommender system are widespread, especially in the e-commerce business. Building a recommender system inside a company is useful but can be difficult. This project-dissertation describes and test different recommender system methods to be implemented inside the company Etilux. The chosen methods are association rule mining, rule-based mining and some collaborative filtering methods such as item-based, user-based collaborative filtering and the graph-model. The main objective is to provide better understanding of the data that can be exploited inside the company as well as key insights on how to reach customers. The results of the analysis will provide a new type of knowledge that the firm will be able to exploit. This paper follows the process of data problem analysis. First, the choice of data and its pre-processing processed are discussed. Then, a presentation of the different methods that will be used is made. Finally, the methods are tested on data from the company and results are presented. Those results will be interpreted and recommendations will be stated.