PopHybrid: a novel item popularity-aware hybrid approach for long-tail recommendation

Creative Commons License

Yalçın E.

4th International Congress on Human-Computer Interaction, Optimization and Robotic Applications, HORA 2022, Ankara, Türkiye, 9 - 11 Haziran 2022 identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/hora55278.2022.9800006
  • Basıldığı Şehir: Ankara
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: collaborative filtering, long-tail recommendations, popularity bias, recommender systems
  • Sivas Cumhuriyet Üniversitesi Adresli: Evet

Özet

© 2022 IEEE.Producing recommendations where tail items have been included more is usually considered one crucial task of recommender systems to improve the diversification ability of the platform. The final recommendations are generated using only a single algorithm in traditional settings; however, algorithms can show varying performances for users differentiating in terms of particular characteristics. Based on this fact, we propose a novel popularity-aware recommendation technique, namely PopHybrid, focusing on increasing the visibility of the items in the long-tail in the final recommendations by hybridizing several algorithms suitably. In particular, this technique chooses the most suitable recommendation output based on specific criteria related to item popularity among multiple algorithms for each user. Thus, it helps control the popularity bias problem and improves the system's overall ability to provide long-tail recommendations. The experiments performed on two popular datasets demonstrate that combining seven collaborative filtering algorithms via our PopHybrid strategy significantly decreases the average popularity of the recommended items and increases the ratio of tail items in the suggested ranked lists.