Integrating Realtime User Behavioral Embeddings Into OpenSearch
summary
Uses OpenSearch User Behavior Insights plus Spark notebooks to ingest clickstreams, train latent behavioral embeddings with alternating least squares, and serve realtime personalized rankings by boosting items scored from collaborative-filtering vectors. Implementation clusters catalog vectors, picks k nearest clusters to the query, and computes a context-aware user vector so searches (example: microwave) surface category-matching items like Samsung stainless steel microwaves instead of unrelated hits. Includes open‑source code and cluster guardrails to limit over‑personalization.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 3:33What are embeddings
- 11:13User intent in search
- 13:18User signals in Realtime
- 15:03OpenSearch setup
- 18:30Personalization spectrum
- 22:02Collaborative filtering
- 26:37Latent features in CF
- 31:09LLM embeddings for personalization
- 33:37Contextual personalization
- 37:52Combining embeddings
- 39:17Pitfalls and future
- 40:18Q&A
related talks
41mText to Context: How We Introduced Hybrid Search
21mPersonalize Search Results with OpenSearch Agentic Memory
26mAdvancing Multi-Modal Search Capabilities in Search Pipeline
31mSponsored Session: From Clicks to Rankings: Building a Learning-to-Rank Pipeline with…
18mAdvancing Search at Uber with OpenSearch
40mAgentic Relevance Tuning With OpenSearch: Faster, Easier, and Higher Quality Search at Scale
46mNatural Language & Search, Improve your Search w/ Vector capabilities
46mDelivering the next wave of innovation with OpenSearch