VCG: A Multimodal Retrieval Framework for E-Commerce Video Feeds under Extreme Cold-Start Conditions

2026-06-17

AI Summary

This work addresses the challenges of user interaction scarcity and positional/duration biases in e-commerce short videos under extreme cold-start scenarios by proposing the first zero-shot multimodal retrieval framework tailored for e-commerce video streams. Built upon a domain-adapted CLIP vision-language model, the framework aligns user intent and video content into a shared semantic space, enabling bidirectional product-to-video and semantic retrieval. It further provides a systematic comparison between discriminative (CLIP-based) and generative (LLM-based) embedding strategies to optimize candidate generation. Online A/B testing demonstrates that the proposed approach significantly improves deep video completion rates by 50%, validating its effectiveness and scalability in large-scale real-world e-commerce environments.

Abstract

The digital commerce landscape is shifting from static, search-driven catalogs to dynamic, immersive video feeds. This transition introduces an ``extreme cold-start'' problem: unlike traditional items, new short-form videos lack the dense interaction history required for collaborative filtering. Furthermore, immersive feeds introduce strong position and duration biases that distort standard engagement signals. In this paper, we demonstrate the Video Candidate Generation (VCG) system, a scalable multimodal retrieval engine designed to solve these challenges in a large-scale e-commerce environment. By leveraging a domain-adapted vision-language model (based on CLIP), we map users and videos into a shared semantic space, enabling zero-shot retrieval based on visual content rather than behavioral history. We detail the system's architecture and present a rigorous evaluation comparing generative (LLM) vs. discriminative (CLIP) embeddings. Our results show that while generative models excel at attribute prediction, they suffer from embedding space collapse in retrieval tasks. Online A/B testing demonstrates that VCG effectively mitigates engagement biases, yielding a 50\% uplift in deep video completion. To showcase the system's capabilities, we present an interactive demonstration featuring three bi-directional retrieval scenarios: Product-to-Video, Video-to-Product, and Zero-Shot Semantic Search.

Problem

• cold-start
• video retrieval
• e-commerce
• engagement bias
• multimodal

Innovation

• multimodal retrieval
• extreme cold-start
• vision-language model
• zero-shot retrieval
• embedding space collapse