Reproduce “Transformers Meet ACT-R: Repeat-Aware and Sequential Listening Session Recommendation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Viet-Anh Tran, Guillaume Salha-Galvan, Bruno Sguerra, Romain Hennequin
A system developed by Deezer Research, PISA, integrates Transformer architecture with ACT-R cognitive principles to provide repeat-aware sequential listening session recommendations. This approach simultaneously captures dynamic preference evolution and accurately models repetitive listening patterns, achieving strong performance on both repeated and novel song suggestions across large-scale music datasets.
Key insights
- Explicitly modeling repetitive patterns through cognitive principles (ACT-R) is crucial for effective music recommendation, leading to improved performance.
- Integrating ACT-R within a dynamic Transformer-based sequential model yields superior results compared to using ACT-R solely for inference or with static models.
- The system can effectively balance recommending both repeated and novel songs, overcoming a limitation of many repeat-aware baselines that neglect exploration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.