Hybrid adaptive methor to predict user preferences

Postgraduate Thesis uoadl:1320500 641 Read counter

Unit:
Κατεύθυνση / ειδίκευση Θεωρητική Πληροφορική (ΘΕΩ)
Library of the School of Science
Deposit date:
2014-04-09
Year:
2014
Author:
Μουζανίδης Αλέξανδρος
Supervisors info:
Ιωάννης Ιωαννίδης Καθηγητής ΕΚΠΑ, Γεώργιος Παλιούρας Ερευνητής Β' ΕΚΕΦΕ "Δημόκριτος"
Original Title:
Υβριδική προσαρμοστική μέθοδος πρόβλεψης μεταβαλλόμενων προτιμήσεων χρηστών
Languages:
Greek
Translated title:
Hybrid adaptive methor to predict user preferences
Summary:
We use the logged interaction between users and the system over the time and we
try to detect what users likes and what is interesting for them. We extract
content based, collaborative based and demographic based user profiles and we
combine these to make estimations about how interesting and significant are the
features of the items to any user. The new algorithm uses the content-based,
collaborative based and demographic based estimations about item features to
predict “rates” for items that are not yet considered. From the log files we
extract the statistical significance about item features and past user rates to
adapt to the current rates that a user would give
Keywords:
Machine Learning, Filtering Methods, User modeling, Interest Drift, Concept Drift
Index:
No
Number of index pages:
0
Contains images:
Yes
Number of references:
33
Number of pages:
59
document.pdf (639 KB) Open in new window