Leveraging In-Context Learning to Transfer Cross-Domain Knowledge in Click-Through Rate Prediction Baglam I i grenme Kullanilarak Tiklama Orani Tahminlerine apraz Alan Bilgilerinin Aktarilmasi

Aydogdu M. E., Sengor Altingovde I., KARAGÖZ P., TOROSLU İ. H.

33rd IEEE Conference on Signal Processing and Communications Applications, SIU 2025, İstanbul, Türkiye, 25 - 28 Haziran 2025, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/siu66497.2025.11112006
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: cross-domain recommendation, CTR, large language models, recommendation
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

With the development of large language models, natural language tasks have seen significant improvements, including personalized recommendation. Traditional approaches in recommendation are often based on collaborative filtering, which relies on historical interactions of similar users. These methods struggle with cold-start and data sparsity issues. Cross-domain recommendation systems try to tackle these problems by leveraging knowledge from a richer domain to increase recommendation performance. However, this is a tedious task as it needs to correlate between two different domain knowledge. Pre-trained LLMs (Large Language Models), on the other hand, can tackle these problems thanks to their parametric knowledge and the ability to generate rich representations of user preferences and contextual information. This article analyzes the use of pre-trained LLMs relying on the parametric knowledge and in-context learning for Click-Through Rate (CTR) prediction.