Extended feature combination model for recommendations in location-based mobile services


Sattari M., TOROSLU İ. H., KARAGÖZ P., Symeonidis P., Manolopoulos Y.

KNOWLEDGE AND INFORMATION SYSTEMS, cilt.44, sa.3, ss.629-661, 2015 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 44 Sayı: 3
  • Basım Tarihi: 2015
  • Doi Numarası: 10.1007/s10115-014-0776-5
  • Dergi Adı: KNOWLEDGE AND INFORMATION SYSTEMS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Sayfa Sayıları: ss.629-661
  • Anahtar Kelimeler: Feature combination, Extended feature combination, Location-based social networks (LBSN), Recommendation, Tensor, Singular value decomposition (SVD), Higher-order singular value decomposition (HOSVD), CONTEXT, INFORMATION, SIMILARITY, WORDNET, SYSTEM
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

With the increasing availability of location-based services, location-based social networks and smart phones, standard rating schema of recommender systems that involve user and item dimensions is extended to three-dimensional (3-D) schema involving context information. Although there are models proposed for dealing with data in this form, the problem of combining it with additional features and constructing a general model suitable for different forms of recommendation system techniques has not been fully explored. This work proposes a technique to reduce 3-D rating data into 2-D for two reasons: employing already developed efficient methods for 2-D on a 3-D data and expanding it with additional features, which are usually 2-D also, if it is necessary. Our experiments show that this reduction is effective. The proposed 2-D model supports content-based, collaborative filtering and hybrid recommendation approaches effectively, whereas we have achieved the best accuracy results for pure collaborative filtering recommendation model. Since our method was built on efficient singular value decomposition-based dimension reduction idea, it also works very efficiently, and in our experiments, we have obtained better run-time results than standard methods developed for 3-D data using higher-order singular value decomposition.