Your browser does not support JavaScript!

Home    Σχετικά με την Ανίχνευση Τάσεων στους Αλγόριθμους Συστάσεων  

Results - Details

Add to Basket
[Add to Basket]
Identifier uch.csd.msc//2007rousidis
Title Σχετικά με την Ανίχνευση Τάσεων στους Αλγόριθμους Συστάσεων
Alternative Title Identifying trends in recommendation algorithms
Author Rousidis, Ioannis
Thesis advisor Πλεξουσάκης, Δημήτρης
Abstract Digital and Social Networking Revolution plays a major role in our everyday life. The social prevalence of this can be evidenced by the evolution of, and demand for, personalized media and on-line shopping. Recommender Systems (RS) are applications that provide personalized advice to users about products or services they might be interested in. However, the ultimate effectiveness of an RS is dependent on factors concerning the underlying algorithm. In this work we report on the algorithmic aspects of the collaborative filtering (CF) method, the prevalent method for providing recommendations. More specifically, we focus on three key factors of the CF: efficacy, efficiency and accommodation to new data. Efficacy involves the quality of the recommendation result. Moreover, especially CF suffers from slow response time, because each single prediction requires the scanning of a whole database of user ratings, making thus efficiency a hot topic. Finally, a RS must be capable of handling new data, be it new users or new items. To provide high-quality recommendations, we develop a theoretical framework for combining information from different sources. Two combinations schemes are being examined: the weighted sum and the product rule. We experimented with our schemes using in each single prediction the following sources: user ratings, item ratings and the categories they belong. Our experimental evaluation indicates that our combination schemes significantly improve the quality performance of the CF method. In order to improve the efficiency and accommodate to new data, we propose the Incremental Trend Diagnosis (ITD), a novel framework for CF. This approach uses dimensionality reduction to create a model on user rating trends which is updated incrementally. CF is applied upon trends for the estimation of user-to-user similarities. Thus, we replace complex similarity estimations with a scalar operation upon rating trends. Experimental results show that our framework is capable of formulating high quality recommendations which makes ITD suitable for online application. To our knowledge, little work has addressed the use of incremental updating model in collaborative filtering.
Language English
Issue date 2007-05-21
Collection   School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses
  Type of Work--Post-graduate theses
Views 435

Digital Documents
No preview available

Download document
View document
Views : 9