Joint Spatio-Temporal Enlargement for Micro-Video Popularity Prediction
A recent study published on arXiv (2604.20311) introduces a comprehensive framework for predicting the popularity of micro-videos (MVPP), enhancing both temporal and spatial dimensions simultaneously. Current MVPP techniques face challenges due to sparse short-range temporal sampling and a limited capacity for retrieval memory. The proposed framework allows for accurate analysis of very long video sequences and features a scalable memory bank capable of infinite expansion to include all pertinent historical information. This advancement seeks to enhance content recommendations and optimize traffic distribution on digital media platforms.
Key facts
- Paper ID: arXiv:2604.20311
- Announce Type: cross
- Focus: micro-video popularity prediction (MVPP)
- Goal: forecast future popularity of videos on online media
- Applications: content recommendation, traffic allocation
- Limitations addressed: sparse short-range temporal sampling, flat retrieval memory with limited capacity
- Proposed solution: joint spatio-temporal enlargement framework
- Enables precise perception of extremely long video sequences
- Supports infinitely expandable scalable memory bank
Entities
Institutions
- arXiv