Abstract
This thesis studied Recommender Systems from a Multiple-Criteria Decision Analysis perspective. It was proved that, two different fields of research, the one of Recommender Systems, a recently developed field of Information Retrieval and the other of Multiple-Criteria Decision Analysis, a mature now field of Decision Science, share common goals and objectives and thus their merging, can be proved extremely dynamic and effective. Nevertheless, careful consideration must be taken to achieve maximum results. Recommender Systems are being developed for about 20 years now and despite their exponential growth, they are still considered in their infancy from a research point of view. This means that yet, several aspects of these systems are to be explored. Chapter 6 Concluding remarks Dept. of Production Eng. & Management Technical University of Crete 154 L‘Universite Paris-Dauphine Recommender Systems grew out of Information Retrieval, to overcome the natural consequence of the information a ...
This thesis studied Recommender Systems from a Multiple-Criteria Decision Analysis perspective. It was proved that, two different fields of research, the one of Recommender Systems, a recently developed field of Information Retrieval and the other of Multiple-Criteria Decision Analysis, a mature now field of Decision Science, share common goals and objectives and thus their merging, can be proved extremely dynamic and effective. Nevertheless, careful consideration must be taken to achieve maximum results. Recommender Systems are being developed for about 20 years now and despite their exponential growth, they are still considered in their infancy from a research point of view. This means that yet, several aspects of these systems are to be explored. Chapter 6 Concluding remarks Dept. of Production Eng. & Management Technical University of Crete 154 L‘Universite Paris-Dauphine Recommender Systems grew out of Information Retrieval, to overcome the natural consequence of the information age, which is called information overload and often leads users of Information Systems to despair and frustration. As a consequence, these systems try to filter out useful information for the respective user. To achieve that, accurate user modeling is undoubtedly considered their most important step and thus user profiling and modeling consist brotherly fields of Recommender Systems. On the other hand, Multiple Criteria Analysis, is extensively studied in Operations Research due to the fact that real world problems are intrinsically multidimensional and many Multiple Criteria Analysis have nowadays developed and successfully applied to several managerial and other problems. Multiple criteria Analysis may be considered as a set of methodologies that process several criteria simultaneously. This means that by default, an MCDA methodology assumes that multiple, often conflicting criteria are involved in a decision making process. To this end, it tries to model decision maker‘s value system by considering all the underlying attributes that lead to a specific decision. It is advocated in this thesis that methodologies from the MCDA field can be proved helpful in solving common problems of Recommender Systems. In particular, fully automated collaborative filtering Recommender Systems suffer from the so called ―cold start? problem. This problem, either concerning new items or new users of the system, is apparent when insufficient information is gathered for this item or user. In UTARec, however, the cold start problem for new items is limited by the fact that, even if only once this item is rated, it enters the system and is ready to be recommended to all users that belong to the same group with the user that provided the initial rating. The way in which users are modeled under the framework of this thesis, enables a less vulnerable to cold start method, since a new user, is assigned in a group and automatically adopts properties of this group and a new item as soon as it is rated by one user it enters the system and is likely to be recommended to many users simultaneously. Moreover, the data sparseness of traditional collaborative filtering Recommender Systems is limited in the case of a UTARec type Recommender Chapter 6 Concluding remarks Dept. of Production Eng. & Management Technical University of Crete 155 L‘Universite Paris-Dauphine
show more