Compressed Video Aggregator: Content-driven Module for Efficient Micro-Video Recommendation

2026-05-09

AI Summary

This work addresses efficiency and performance bottlenecks in micro-video recommendation caused by information redundancy, high computational overhead, and coarse frame sampling. To this end, the authors propose a lightweight, content-driven module that decouples video representation from user preference learning. The method leverages embeddings from a frozen video foundation model (VFM) to perform implicit inference without cross-attention mechanisms and introduces a title-guided CLIP-based keyframe resampling strategy to produce compact yet semantically rich video representations. Evaluated on the MicroLens and Short-Video datasets, the approach significantly improves recommendation accuracy while reducing training time and GPU memory consumption by several orders of magnitude. Moreover, the proposed frame resampling strategy effectively enhances the performance of existing recommendation models when integrated as a plug-in component.

Abstract

We propose Compressed Video Aggregator (CVA), a lightweight micro-video recommendation module that decouples video information from preference learning. It aggregates frozen VFM embeddings, and uses latent reasoning without cross-attention projection, producing compact video embeddings for recommenders. Due to the redundancy in the frame count of the original benchmark and its overly coarse sampling, we used titles to re-select key frames based on CLIP. Experiments on MicroLens and Short-Video show consistent gains with orders-of-magnitude reductions in training time and GPU memory, and re-selected frames can further enhance the performance of all methods, including CVA. Furthermore, we also discussed the impact of several scenarios involving erroneous titles on our method. Code will be released soon.

Problem

• micro-video recommendation
• video redundancy
• frame sampling
• computational efficiency
• content-driven recommendation

Innovation

• Compressed Video Aggregator
• micro-video recommendation
• frame selection
• latent reasoning
• CLIP