Browsing by Author "Işık, Harun"

Now showing 1 - 1 of 1

Content based user preference modeling for image recommender systems
(Bahçeşehir Üniversitesi Fen Bilimleri Enstitüsü, 2015-01) Işık, Harun; Özden, Kemal Egemen
This thesis deals with evaluating image descriptors on whether they are useful to create a user preference model about user’s taste on images and also whether these models can eventually be used in image recommender systems. Our aim is to address a simple user preference vector by using many visual descriptors of images. By means of image descriptors, we can reveal a correlation between user’s taste and image features and easily build up a vector that models user’s preferences. This content-based relationship may be used for image recommendation. Recommender systems can generally be considered as two headings such as content-based approaches and collaborative filtering approaches. Typical content-based methods computes content in user preference and compare it with other items. We want to use our image descriptor correlation as a content-based approach. But there are some natural challenges about this type contentbased algorithm. For a very large image dataset, computing pairwise distances between vectors of image descriptors is very exhaustive process. To overcome this complexity, we have proposed a novel approach that we make cluster dataset through image feature vectors. This technique may be useful in different ways such that it speeds up image matching since you do not have to match each candidate against each image that a user likes. Also it can be able to group images very meaningfully in term of semantic according to your clustering algorithm success.