Case Study: Dailymotion boosts video recommendations with Qdrant Vector Search

Dailymotion's Journey to Crafting the Ultimate Content-Driven Video Recommendation Engine with Qdrant Vector Database

Dailymotion, a major video-sharing platform, faced significant challenges in providing high-quality, real-time video recommendations. Their traditional collaborative filtering model tended to favor only popular content, leaving new and niche videos undiscovered. They needed a content-based system that could process embeddings for over 420 million videos, handle 2,000 new uploads every hour, and deliver recommendations in under 100ms. To overcome this, they turned to the vendor Qdrant for its vector search capabilities.

By implementing Qdrant's vector database, Dailymotion built a new content-based recommendation engine. The solution processes video titles, tags, descriptions, and transcripts to generate embeddings, which Qdrant uses to perform fast and accurate similarity searches. This integration drastically reduced the processing time for new video content from five hours to just a few minutes. The results were substantial, leading to a more than threefold increase in clicks and user interactions with recommended videos, while also effectively solving the cold-start problem for new content.

Dailymotion

Gladys Roch

Machine Learning Engineer