Λειτουργίες σύγκρισης και πράξεων με ανωνυμοποιημένα αριθμητικά δεδομένα
Comparison and operations with anonymized numerical data
View/
Keywords
Αλγόριθμοι εξατομίκευσης ; Φιλτράρισμα βάσει περιεχομένου ; Συνεργατικό φιλτράρισμα ; Παραγοντοποίηση πινάκων ; ML.NETAbstract
The rapid growth of digital content has made it increasingly challenging for users to find information that truly matches their needs and preferences. In this context, recommender systems have emerged as valuable tools, offering personalized suggestions and helping users navigate environments where the sheer amount of information can be overwhelming.
This thesis focuses on the study, implementation, and evaluation of three different approaches to personalized recommender systems. The first is a content-based filtering system developed in Python, which uses techniques such as shingling, MinHash signatures, and Locality Sensitive Hashing to detect similarities between service descriptions. In addition, a collaborative filtering system was also built in Python, relying on user ratings to suggest new services by considering the preferences of users with similar interests. Finally, a matrix factorization–based system was implemented in C# using Microsoft’s ML.NET library, trained on a large and sparse dataset.
The methodology combined both the theoretical analysis of each technique and the development of working code. For each implementation, the challenges encountered were explored along with the potential of each method to address them.
Department
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα ΠληροφορικήςNumber of pages
53Language
GreekCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα