- Vector similarity search and Two-Tower model: How Google finds valuable data in milliseconds
- Googleを支えるベクトル近傍検索技術とVertex Matching Engine
45 min
Data analysts, data scientists and DevOps engineers. Intermediate level
How does Google find valuable data in milliseconds from the vast sea of the Internet contents? Vector similarity search is the core information retrival technology that realizes the user experience of Google's core services including Google Search, YouTube, Play and many others. Another key component is the Two-Tower model that learns the relationships from pairs of query and candidate features, extending the capability of the information retrieval far from the traditional recommendation algorithms. As Google Cloud released those as a commercial service, businesses are starting to incorporate them for transforming their services. We will learn the technology details and how users are applying them in the real-world solutions.
See also:
- Find anything blazingly fast with Google's vector search technology
- Mercari leverages Google's vector search technology to create a new marketplace
ベクトル近傍検索は、Google検索をはじめ、Google画像検索、YouTubeの動画の推薦、Google Playストア等で幅広く利用されている「Googleを支える技術」です。この大規模かつ低遅延の検索バックエンドを利用するVertex Matching Engineについて紹介し、情報検索やリコメン、分類等々、さまざまな用途への適用について解説します。