Closing the Long-Short View Gap in Sequential Recommendation Without Cached History
_reachsumit · x · 2026-09-09
A CIKM 2026 paper (Lingfeng Shi et al.) tackles the long-short view gap in sequential recommenders: models trained on long user histories degrade sharply when serving only recent behaviors due to real-time efficiency limits, and prior history-compression approaches add infrastructure overhead and fail in cold-start scenarios.
The authors identify structural flaws rooted in embedding geometry and dataset sparsity, then propose a two-stage framework: replacing dot-product scoring with angular similarity plus a modified softmax to counter prefix position bias, followed by fine-tuning only universal bias and LayerNorm components. Experiments across two backbones and three public datasets show the gap can be closed without caching per-user history states — a lightweight deployment path for practitioners.
More from Research
- Jev's eval abstraction maps 1:1 to autorubric paper from 8 months ago, researcher finds — deliprao · 2026-09-21
- Researcher Says TypeSafe's Jev Mirrors His Autorubric LLM Eval Framework From 8 Months Ago — deliprao · 2026-09-21
- As ICLR 2027 Tops 60K Submissions, Researcher Proposes 3-4 Paper Cap per Author — ziv_ravid · 2026-09-21
- ICLR 2027 Hits 60K+ Submissions; Researcher Proposes Paper Caps, Forced Reproducibility — ziv_ravid · 2026-09-21
- No, Laya isn't capped at 512 tokens — it's ModernBERT with 8192-token configs — antoine_chaffin · 2026-09-21
- Kev open-source decision models scale to 0.6B/4B/8B, trainable in 40 min on one H100 — TheMoonMidas · 2026-09-21