National and Kapodistrian University of Athens
Mining and managing user-generated content and preferences
Abstract
dc:descriptionThe World Wide Web is evolving, as new technologies constantly emerge. Capturing the users’ preferences is essential to provide them with engaging services that better fulfill their needs. For this reason, this thesis proposes distinct ways to consider user preferences, either in pre-usage or post-usage scenarios. Skyline queries are a characteristic example where the user may specify her preferences in a pre-usage scenario. A problem of skyline queries is that the result size may be too large. To counter this problem, we propose techniques to diversify the result of a skyline query. Diversifying the skyline is mapped to an NP-Hard problem (k-dispersion). Therefore we propose efficient approximations to solve it. As a post-usage scenario, users are able to provide feedback through online reviews. These reviews describe the experience of a user with a product / service, and what aspects / features of the product they liked or did not like. Using this information, we present techniques to identify competitive products. Competitive products are those that can cover similar needs of similar groups of people. We performed an extensive evaluation of our methodology and demonstrated that our model is able to efficiently and effectively capture the competitiveness of items. Finally, user preferences and interests may be captured through information posted online, in social media platforms. Social media data can be used to identify events that occur in the physical world. Unfortunately, given the particularities of social media, straightforward solutions (namely online clustering) do not perform well. To address this problem, we resorted to affective theories of emotion from psychology, according to which external stimuli (i.e. events) have an effect on a person’s emotional state and compel them to externalize their thoughts. Using these theories as our starting point, we monitored the aggregate emotional state of large groups of people, geographically dispersed, and automatically identified abrupt changes. The assumption is that abrupt changes in the aggregate emotional state of the group is due to external stimuli. To achieve this, we developed custom geocoding techniques that convert text to GPS coordinates, a classification framework mapping social media data to emotions, an online outlier detection algorithm that operates on the probability density function of the aggregate emotional state, and a visualization technique to ease the presentation of events.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2014
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Valkanas, Georgios
- Βαλκανάς, Γεώργιος
Subjects
dc:subject × 14- Εξόρυξη δεδομένων στο διαδίκτυο
- Προτιμήσεις χρηστών
- Εξόρυξη δεδομένων από κριτικές
- Ανάλυση κοινωνικών μέσων
- Αναγνώριση συμβάντων
- Web mining
- User preferences
- Review mining
- Social media analysis
- Event detection
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/42627