Native Dense Vector Scoring in Elasticsearch
summary
Elasticsearch stores ordered numeric embeddings as a dense_vector type since 7.0 and adds built-in scoring functions in 7.3 so dot‑product and cosineSimilarity run inside the engine without a custom plugin. The demo loads MovieLens-style matrices into Spark, uses matrix factorization to produce user/item vectors, indexes them as vectors, and issues script score queries (dot product / top‑k) to return recommended movies — a working end‑to‑end recommender with no cluster plugins.
AI-written summary · about
chapters (4)
- 0:00Introduction
- 0:56Vectors and search relevance
- 4:51Native vector support in ES
- 6:37Implementation and conclusion
related talks
38mSearch: A New Era
27mMachine Learning for Advanced AI Search
47mBuilding intelligent analytics at scale
45mRevolutionizing Search with Power of Neural Sparse and Dense Vectors
01h 23mVector Databases Demystified: Discover how they work
59mContextual search with vector search: exploring your options with open source tools
01h 01mWord Embeddings - An Alternative & Efficient Approach to Search for Docs
28mA Deep Dive into Python-Powered Precision and Scalability