A multi-objective recommendation system


Tezin Türü: Doktora

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, Bilgisayar Mühendisliği Bölümü, Türkiye

Tezin Onay Tarihi: 2016

Öğrenci: MAKBULE GÜLÇİN ÖZSOY

Danışman: FARUK POLAT

Özet:

Recommendation systems suggest items to the user by estimating their preferences. Most of the recommendation systems are based on single criterion, such that they evaluate items based on overall rating. In order to give more accurate recommendations, a recommendation system can take advantage of considering multiple criteria. Beside combining multiple criteria from a single data source, multiple criteria from multiple data sources can be combined. Recommendation methods can also be used in various application domains involving big data such as marketing, biology, chemistry. In this thesis, four applications are studied: 1) use of multiple criteria from a single source to make recommendations, 2) use of multiple criteria from multiple sources to make recommendations, 3) use of recommendation methods to predict gene regularity networks and 4) use of recommendation methods to identify new indicators for known drugs. Firstly, we propose a new multi-objective optimization based recommendation method that combines multiple criteria, namely past preferences of users, hometown of users, friendship relation among users, check-in time information. We expanded this method by inferring home/center location of users in terms of longitude-latitude pairs, by making dynamic recommendations based on temporal preferences of users and by clustering users by their hometown and friendship relations. These methods are evaluated on a Foursquare check-in dataset. Secondly, we combine information collected from multiple different social networks to create integrated models of individuals and to make recommendations to them. To our knowledge, this is the first work aiming to use information from multiple social networks in recommendation process by modeling the users. For this purpose, we collect and anonymize two data-sets that contain information from BlogCatalog, Twitter, Flickr, Facebook, YouTube and LastFm web-sites. We implement several different types of recommendation methodologies to observe their performance while using single or multiple features from a single source or multiple sources. Thirdly, observing the common features of recommendation systems and gene regularity networks (GRNs), we use the proposed multi-objective optimization based recommendation method to predict the gene relationships; such that which genes regulates the others. Lastly, we adapt the proposed recommendation method to identify new indications for known drugs, i.e. drug repositioning.