Unit:
Department of Informatics and TelecommunicationsΠληροφορική
Author:
KOLIOPOULOU KONSTANTINA-ELEFTHERIA
Supervisors info:
Ιωάννης Εμίρης, Καθηγητής, Τμήμα Πληροφορικής και Τηλεπικοινωνιών, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Ιωάννης Χαμόδρακας, Μέλος του Εργαστηριακού Διδακτικού Προσωπικού, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Original Title:
Υλοποίηση και βελτίωση αλγορίθμων διάχυσης σε συστήματα προτάσεων
Translated title:
Implementation and improvement of diffusion algorithms in recommendation systems
Summary:
Due to the explosive growth of the amount of available information on the Internet, the need for management and transmission to the appropriate audience, namely the role of recommendation systems, becomes consequently critical. In this context, various kinds of algorithms have been designed the last years aiming to improve their performance and efficiency. This thesis focuses on the development of recommendation systems with the use of diffusion based algorithms which use bipartite networks. The primary goals of this thesis are the implementation, evaluation and comparison of Weighted Slope One (WSO), Heat Spreading (HeatS) and Probabilistic Spreading (ProbS) algorithms. Initially, the function of each algorithm is explained and the way with which they handle the information given. Afterwards, the Hybrid Spreading algorithm (HybridS) is examined which is a combination of HeatS and ProbS algorithms aiming to produce improved recommendations. Also, the nearest neighbor (locality-sensitive hashing) LSH-Superbit algorithm is used in order to reduce the time complexity of the problem. Last but not least, algorithms are evaluated in terms of efficiency, effectiveness and validity using various known evaluation metrics.
Main subject category:
Algorithms and Theory of Computation
Keywords:
Diffusion based algorithms, bipartite network, Weighted Slope One, HeatS, ProbS, predictions, evaluation metrics, efficiency, effectiveness, validity.
