Diverse but Relevant Recommendations with Continuous Ant Colony Optimization


Yılmazer H., Özel S. A.

Mathematics, cilt.12, sa.16, 2024 (SCI-Expanded) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 12 Sayı: 16
  • Basım Tarihi: 2024
  • Doi Numarası: 10.3390/math12162497
  • Dergi Adı: Mathematics
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Communication Abstracts, Metadex, zbMATH, Directory of Open Access Journals, Civil Engineering Abstracts
  • Anahtar Kelimeler: cold start, collaborative filtering, continuous ant colony optimization, long-tail, recommender systems, top-N recommendation
  • Çukurova Üniversitesi Adresli: Evet

Özet

This paper introduces a novel method called AcoRec, which employs an enhanced version of Continuous Ant Colony Optimization for hyper-parameter adjustment and integrates a non-deterministic model to generate diverse recommendation lists. AcoRec is designed for cold-start users and long-tail item recommendations by leveraging implicit data from collaborative filtering techniques. Continuous Ant Colony Optimization is revisited with the convenience and flexibility of deep learning solid methods and extended within the AcoRec model. The approach computes stochastic variations of item probability values based on the initial predictions derived from a selected item-similarity model. The structure of the AcoRec model enables efficient handling of high-dimensional data while maintaining an effective balance between diversity and high recall, leading to recommendation lists that are both varied and highly relevant to user tastes. Our results demonstrate that AcoRec outperforms existing state-of-the-art methods, including two random-walk models, a graph-based approach, a well-known vanilla autoencoder model, an ACO-based model, and baseline models with related similarity measures, across various evaluation scenarios. These evaluations employ well-known metrics to assess the quality of top-N recommendation lists, using popular datasets including MovieLens, Pinterest, and Netflix.