Kuaishou open-sources KuaiLive-M3, a 35M-interaction dataset for live-stream recommendation
_reachsumit · x · 2026-07-29
Kuaishou released KuaiLive-M3, a multi-modal, multi-domain, multi-feedback dataset for live-stream recommendation.
- The authors argue existing public datasets miss three things: temporally evolving multimodal content, cross-domain user behavior between short video and live streaming, and explicit feedback beyond implicit clicks.
- KuaiLive-M3 includes 21,938 users, 35 million live-stream interactions, and 111 million short-video interactions.
- It also provides around 88 million timestamped segment-level multimodal embeddings to capture content evolution over time.
- The dataset adds 25,403 questionnaire-based feedback records to connect implicit behavior with explicit user preference.
- The post positions the release as a richer benchmark for live-stream recommendation research.
More from Research
- DexRobotics open-sources a 50-episode SO101 robot fine-tuning workflow — AdinaYakup · 2026-07-29
- SmartPhotoCrafter lets a model diagnose a photo first, then auto-edit it — 量子位 · 2026-07-29
- A slide argues Bayes is only one lens, not the theory of modern AI inference — davidmanheim · 2026-07-29
- Brain-inspired AI can plan and solve problems with far less energy, study suggests — striketheviol · 2026-07-29
- From GPT-2 to KimiK3, a thread argues the story is bigger than scale — algo_diver · 2026-07-29
- Synthetic training data helps AI resolve messy dataset citations — RexDouglass · 2026-07-29