SocRipple: A Two-Stage Framework for Cold-Start Video Recommendations

2025-08-10

AI Summary

To address the cold-start problem in video recommendation—where newly uploaded videos lack sufficient user interaction history for personalized distribution—this paper proposes a two-stage retrieval framework leveraging social propagation. In the first stage, an initial exposure is achieved by exploiting the social graph structure. In the second stage, personalized re-ranking is performed by jointly encoding users’ historical behavior embeddings and early sparse interaction signals, followed by a KNN-based diffusion strategy over the user–video interaction graph. The key innovation lies in integrating social propagation mechanisms with graph diffusion modeling to jointly optimize exposure efficiency and personalization accuracy—even in the absence of prior interaction data. Extensive experiments on a large-scale industrial video platform demonstrate that the proposed method increases cold-start video distribution volume by 36% while maintaining stable user engagement rates, significantly outperforming existing cold-start recommendation baselines.

Technology Category

• Computer Vision: Image and Video Retrieval
• Search and Optimization: Distributed Search
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• User Modeling, Personalization and Recommendation: Social recommender systems and personalization for the social Web
• Web Mining and Content Analysis: Content-based information diffusion
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Most industry scale recommender systems face critical cold start challenges new items lack interaction history, making it difficult to distribute them in a personalized manner. Standard collaborative filtering models underperform due to sparse engagement signals, while content only approaches lack user specific relevance. We propose SocRipple, a novel two stage retrieval framework tailored for coldstart item distribution in social graph based platforms. Stage 1 leverages the creators social connections for targeted initial exposure. Stage 2 builds on early engagement signals and stable user embeddings learned from historical interactions to "ripple" outwards via K Nearest Neighbor (KNN) search. Large scale experiments on a major video platform show that SocRipple boosts cold start item distribution by +36% while maintaining user engagement rate on cold start items, effectively balancing new item exposure with personalized recommendations.

Problem

• Addresses cold-start challenges in video recommendations
• Improves personalized distribution of new items
• Balances new item exposure with user engagement

Innovation

• Two-stage retrieval framework for cold-start
• Leverages creator social connections initially
• Uses KNN search with engagement signals