Εφαρμογές τεχνικών μηχανικής μάθησης σε δεδομένα γράφων

Λουρίδας, Σάββας

Master Thesis

Author

Λουρίδας, Σάββας

Date

2024-02

Abstract

In our modern age recommender systems rely heavily on machine learning to automatically learn and improve over time based on user data and feedback. Many websites and applications such as social networks, content services, e-commerce platforms and others use recommendation algorithms to help users find new goods, services or information that may be of interest to them. In order to model a user's tastes and interests, recommender systems use machine learning algorithms to evaluate vast amounts of user data, including ratings, search queries and past purchases. Then, using this model, personalized recommendations are generated based on each user's requirements and tastes. In this master's thesis we aim to study the theory of graphs, meaning the types of graphs, the way they are visualized, applied and the problems during their application. Also we examine in machine learning its forms, its models, its applications and its limitations. Afterwards we proceeded to a practical experiment. For our experiment our methodology was to take the MovieLens 100K dataset, explore it by making useful graph data analysis inferences, bring it into the appropriate heterogeneous bidirectional graph format which we eventually used to train a graph neural network algorithm recommendation system. Finally we did two experiments changing the training criteria and measured the performance of the algorithm in each case with the metrics binary cross entropy loss with logits (loss) and area under the ROC curve (AUC) which we visualized in graphs for each case to make it easier to see the results. In the end we concluded in both cases that our system performed "outstanding".

Postgraduate Studies Programme

Πληροφοριακά Συστήματα και Υπηρεσίες

Department

Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών Συστημάτων

Number of pages

Language

Greek

URI

https://dione.lib.unipi.gr/xmlui/handle/unipi/16285
http://dx.doi.org/10.26267/unipi_dione/3707

Collections

Τμήμα Ψηφιακών Συστημάτων

Show full item record

Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα