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

TL;DR

SocRipple is a two-stage framework that improves cold-start video recommendations by leveraging social connections and early engagement signals, significantly increasing new item exposure without sacrificing user engagement.

Contribution

It introduces a novel two-stage retrieval framework combining social graph data and engagement signals for better cold-start item distribution.

Findings

Boosts cold start item distribution by +36%

Maintains user engagement rate on cold start items

Effective balance between new item exposure and personalization

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.