Back to search

Aristotle University Of Thessaloniki (AUTH)

Τεχνικές εξόρυξης γνώσης για χαρακτηρισμό περιεχομένου και παροχή συστάσεων στον κοινωνικό ιστό

Abstract

dc:description

The thesis consists of two research sections: the characterization of social content and the delivery of recommendation services to the users of the Social Web. Each section approaches distinctively the analysis of great volumes of data aiming at highlight the most valuable parts according to the application at hand.In the first section, we propose a clustering framework which models images with similarity graphs and clusters them by taking into account both their visual content and the available metadata. Then, we make use of the clustering results along with machine learning algorithms to classify the image clusters into Points of Interests or Events. The classification results are included in a web-based application which enables users to navigate through large collections of photos. Furthermore, regarding the discovery of event-related photos, we present an efficient method for handling great number of photos. The first section is concluded with two more techniques for the characterization of social content. In the first one, we develop a collaborative event annotation framework which approaches the “Event” concept from different perspectives by proposing a series of search functionalities in the presence of large photo collections. The second technique presents a number of matching methods aiming at relating single photos with points of interests. All methods are compared with the search mechanism of Flickr in terms of precision and recall.The second section focuses on the elicitation of user preferences and the delivery of recommendations. We propose two ways of incorporating the notion of Dissimilarity into recommender systems: (1) a dissimilarity recommender and (2) a framework for extracting dissimilarity features of rated items. The dissimilarity recommender is based on calculated differences in users’ preferences among item attributes to extract estimations. The results of the dissimilarity algorithm are combined with the ones coming from existed recommenders based on the notion of similarity. Apart from the aforementioned recommender, the notion of dissimilarity is presented through a feature extraction framework which aims at enriching the representations of items with new indicative features. Each feature refers to an item and expresses the difference in preferences for an attribute between the user who rated it and a group of users who have similar preferences with him. The new features are incorporated into existed recommenders-classifiers and are evaluated in terms of precision. The results have shown that the dissimilarity features can lead to more appropriate recommendations. The second section is concluded with two parallel/distributed solutions for recommender systems. In the first solution, we present IANOS framework through which we can execute similarity recommenders in a parallel mode in order to be combined with our dissimilarity recommender, the procedures of which are developed in a distributed manner. In the second solution, we propose Hydra framework through which we can create weighted-hybrid recommenders by combining any type of existed ones. Hydra’s combination policy (Virtual Fusion) is applied upon the results of the incorporated recommenders by taking into account their individual performance aiming at providing better recommendations.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2013

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Zigkolis, Christos
  • Ζιγκόλης, Χρήστος

Subjects

dc:subject × 14

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/36727

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Zigkolis, Christos; Ζιγκόλης, Χρήστος. Τεχνικές εξόρυξης γνώσης για χαρακτηρισμό περιεχομένου και παροχή συστάσεων στον κοινωνικό ιστό. Aristotle University Of Thessaloniki (AUTH), 2013. http://hdl.handle.net/10442/hedi/36727