Case Study: BIGO improves video deduplication throughput and search speed with Milvus from Zilliz

How Milvus Transformed BIGO's Video Deduplication System for Optimal Throughput and User Experience

BIGO, a global technology company with popular short-video platform Likee, faced a significant challenge in managing user experience due to a massive volume of duplicate videos being uploaded. Their previous solution, Faiss, struggled with the scale of over 700 million vectors, resulting in slow search response and limited throughput. They needed a more efficient and scalable solution for video deduplication.

The company implemented Zilliz's vector database, Milvus, to power its video similarity search system. The solution processes new videos into feature vectors and uses Milvus to perform lightning-fast searches, identifying duplicates in under 200ms with a high recall rate. This resulted in significantly increased query throughput without any performance compromise, fueling the growth of BIGO's short-video business.

BIGO

Xinyang Guo

Software Engineer