Personalized Car Configuration Recommendation System based on Machine Learning

The internet provides companies with the opportunity to create cost efficient one-to-one relationships with their customers. Not only do these customers get more involved with the corresponding website or service but they typically also benefit from an improved user experience. Even though companies and customers can both benefit from these relationships, there are web services which still do not utilize them. Some car configurators for example offer hundreds of configuration components which makes the configuration process complex and time consuming. Personalization as one form of one-to-one marketing has already solved similar problems in other domains. For this reason, the goal of this thesis is to develop, implement and evaluate a concept for applying personalization to the Mercedes-Benz car configurator. The personalization approaches which have already been successfully applied in other domains are used as foundation for the developed concept. The core idea is to recommend a small amount of components which are of interest to the individual user. The recommendations are generated by combining collaborative filtering and machine learning techniques. The result of the thesis is a prototype which implements the developed concept. A conducted user study indicates that users generally appreciate recommendations when configuring a car and that the prototype offers added value over the current Mercedes-Benz configurator. Furthermore, the recommendations themselves mostly match the users’ preferences. With this feedback it can be presumed that the personal ization approach works and that it improves the user experience.