Idea
A real-time interactive dataset and platform enabling live streaming recommendation models for streaming services and AI researchers.
Research Paper
Core Innovation
This paper introduces KuaiLive, the first dataset capturing real-time, multi-behavior user interactions in live streaming with precise timestamps and rich features. It enables realistic simulation of dynamic recommendation scenarios unlike prior static or offline datasets. The dataset supports diverse tasks including multi-task learning and fairness-aware recommendation, advancing live streaming recommendation research.
Market Size (TAM)
$2–10B TAM, $1–2B SAM; assumption: growing global live streaming market and increasing demand for personalized recommendations.
Potential Customers & Pain Points
- Live Streaming Platforms Needing Realistic Recommendation Data
- AI Researchers Lacking Dynamic Interaction Datasets
- Recommendation System Developers Seeking Multi-behavior Modeling Data
Business Model
Subscription-based API access to the dataset and platform for research and commercial recommendation model development; custom data solutions for enterprise clients.
Competitive Landscape
- Twitch Data APIs
- YouTube Live Analytics
- Douyin Live Data Services
Implementation Challenges
- Data Privacy and Compliance Challenges
- High Complexity of Real-time Data Processing
- Integration with Existing Recommendation Systems
Validation Strategy
- Pilot integration with select live streaming platforms for recommendation improvement
- Benchmark studies comparing model performance using KuaiLive versus existing datasets
- User feedback collection from developers and researchers using the dataset
Research Paper Overview
KuaiLive: A Real-time Interactive Dataset for Live Streaming Recommendation
Summary
KuaiLive is the first real-time, interactive dataset from Kuaishou, a major Chinese live streaming platform, capturing 21 days of user-streamer interactions including clicks, comments, likes, and gifts. It provides precise live session timestamps and rich user and streamer features, enabling realistic simulation and modeling of dynamic live streaming recommendation scenarios. The dataset supports tasks like top-K recommendation, click-through rate prediction, watch time prediction, gift price prediction, multi-behavior modeling, multi-task learning, and fairness-aware recommendation.