Συντάχθηκε 01-02-2010 12:34
από Thekla Papadaki
Email συντάκτη: tpapadaki<στο>tuc.gr
Ενημερώθηκε:
-
Ιδιότητα: σύνταξη/αποχώρηση υπάλληλος ΜΠΔ.
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Τμήμα Μηχανικών Παραγωγής και Διοίκησης
Παρουσίαση Διδακτορικής Διατριβής
Λακιωτάκη Κλεάνθης
Δευτέρα 8 Φεβρουαρίου 2010, ώρα 13.00, Αίθουσα Συνεδριάσεων ΜΠΔ
«An integrated Recommender System based on Multi-Criteria Decision Analysis and Data Analysis methods: Methodology, implementation and evaluation»
Εξεταστική επιτροπή:
1. Ματσατσίνης Νικόλαος, Καθηγητής, ΜΠΔ, Πολυτεχνείο Κρήτης (επιβλέπων)
2. Τσουκιάς Αλέξης, Διευθυντής έρευνας CNRS - Université Paris Dauphine (μέλος τριμελούς)
3. Δουλάμης Αναστάσιος, Επίκουρος Καθηγητής, ΜΠΔ, Πολυτεχνείο Κρήτης (μέλος τριμελούς)
4. Πάσχος Ευάγγελος, Καθηγητής, LAMSADE - Université Paris Dauphine
5. Μυγδαλάς Αθανάσιος, Καθηγητής, Γενικό Τμήμα της Πολυτεχνικής Σχολής, Αριστοτέλειο Πανεπιστήμιο Θεσσαλονίκης
6. Βλαχοπούλου Μάρω, Καθηγήτρια, Τμήμα Εφαρμοσμένης Πληροφορικής, Πανεπιστήμιο Μακεδονίας
7. Μάνθου – Φραγκοπούλου Βασιλική, Καθηγήτρια, Τμήμα Εφαρμοσμένης Πληροφορικής, Πανεπιστήμιο Μακεδονίας
Περίληψη
With the growth of the available information on the Web, the diversity of its users and the complexity of Web applications, researchers started to question this generic approach of “one size fits all”. Does it make sense for an e-commerce Web site for example, to present the same products to internet users with widely diverse preferences?
To address this kind of questions, researchers started developing adaptive Web systems that tailored their appearance and behavior to each individual user or user group. Recommender systems assist search and browsing based information tasks, by recommending items that seem most relevant to users’ interests and might otherwise be missed due to information overload. These 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.
This thesis aims at providing new insights to the development of Recommender Systems, by introducing the exploitation of methodologies and techniques from the Multiple Criteria Decision Analysis (MCDA) field to the Recommender Systems research field. A hybrid methodological framework, which merges techniques from the two aforementioned research areas, is described and analyzed in details. The latter, which also constitutes the major outcome of this thesis, is implemented through UTARec, a system that incorporates the proposed framework and demonstrates its performance. The contribution of this work lies mainly in the potentiality of the proposed hybrid methodological framework and can be divided based on the various disciplines that are benefitted from the results of this thesis, such as User Modeling, Recommender Systems, Multiple Criteria Decision Analysis and e-marketing. A thematic reference of the individual components that build the overall contribution of this work can be found in section 1.2.
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, some of the major shortcomings of current collaborative filtering Recommender Systems such as the so called “cold start” problem, the data sparseness or the unusual rater problem, are limited in the case of UTARec type Recommender Systems. Moreover, common problems of existing content based Recommender Systems such as the feature extraction dependence are also addressed at some point by systems designed according to the proposed methodology. Analytical details on how these problems are addressed are found throughout the individual chapter conclusions and they are also summarized in section 6.1. This thesis ends, with a reference on possible future aspects of this work, providing thus ideas for further research.
