ClockRoPE enhances LLMs for temporal routine modeling · arXiv research

Researchers have developed ClockRoPE, a novel method for temporal routine modeling that enhances the performance of transformer-based large language models, particularly in sequential recommendation tasks. This new approach utilizes Random Fourier Rotations, derived from the Fourier transform of attention modulation functions, to better capture complex distance-correlation patterns like temporal periodicity. ClockRoPE has shown consistent improvements in engagement metrics during online A/B tests and is already in production use within a large video-sharing platform's generative retrieval system. AI

IMPACT This research could improve the performance of recommendation systems and generative retrieval by better modeling temporal patterns in user behavior.

RANK_REASON The item is an arXiv preprint detailing a new method for temporal routine modeling. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• ClockRoPE
• Fourier transform
• online video platform
• Random Fourier Rotations
• Rope
• Rotary Position Embedding
• Sequential Recommendation via Cross-Domain Novelty Seeking Trait Mining

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →