| Περίληψη: | Recommender Systems were developed to address the information overload problem resulted from the rapidly increasing use of the Internet, in order to meet the individual needs of each user. Although having a wide scope of application, they are mainly used in the field of e-commerce to provide the user with the opportunity to easily find the items that interest him the most, which benefits both the companies and the customers. The main objective of this thesis is to develop models, based on machine learning algorithms, that generate recommendations. At first, the theoretical background of Recommender Systems is described and the most well-known approaches for generating personalized recommendations are presented. The basic data mining procedures used in this context are analyzed and then applied to our own dataset, which consists of user ratings for various hotel units. The experimental analysis is divided into two cases of recommendations. In the first one, the goal is to predict the rating a user would give to a hotel, and in accordance we decide whether a recommendation of the given hotel will be made or not. In the second one, we try to directly predict the hotel that a user will visit. A detailed approach on how to address the recommendation problem in either case, along with the arising challenges are presented. For the development of these models, supervised learning techniques are proposed and more specifically classification algorithms are engaged. Finally, the predictions of each classification algorithm, for the two cases, are evaluated and the final conclusions are drawn, followed by suggestions for possible modifications in future work.
|
|---|
