Databricks 矢量搜索：内容、原因和方式（重复）.pdf

1、2024 Databricks Inc.All rights reserved1MOSAIC AI MOSAIC AI VECTOR SEARCH:VECTOR SEARCH:WHAT,WHY AND WHAT,WHY AND HOWHOWSonali Guleria,Ankit VijSonali Guleria,Ankit Vij06/13/2406/13/242024 Databricks Inc.All rights reserved“Ever tried finding a slackEver tried finding a slackmessage but couldnt reme

2、mber message but couldnt remember the exact terms?“the exact terms?“22024 Databricks Inc.All rights reservedIntroduction to Vector Search When not to use?What is Mosaic AI Vector SearchBest Practices for Path to ProductionData IngestionVector Search RetrievalConclusionQ/AAgendaAgenda32024 Databricks

3、 Inc.All rights reserved4Vector SearchVector SearchSearch with understandingSearch with understandingSource:https:/www.cs.cmu.edu/dst/WordEmbeddingDemo/tutorial.htmlSearch based on semantic closeness or inherent characteristics using embeddingsDataVector embeddingsEmbedding model2024 Databricks Inc.

4、All rights reservedRAG systems like chatbotsRecommendation systemsAnomaly DetectionImage and Video RecognitionDrug Discovery“Classic”use-cases such as Personalization,product searchand so on5Vector SearchVector SearchUse CasesUse Cases2024 Databricks Inc.All rights reservedStandard Database Look-upC

5、onventional keyword searchMetricsAggregationsOther ObservabilityWhat it is notWhat it is not6Not an exact matchNot an exact match2024 Databricks Inc.All rights reservedFast and Fault TolerantReal-time Updates and Change DetectionEffective Data PreprocessingHighly ScalableQuality results with high Re

6、callLow LatencyEffectively use the Metadata as filtersAdditional capabilities like Hybrid searchIngestionIngestionRetrievalRetrieval7Main Components for Vector SearchMain Components for Vector Search2024 Databricks Inc.All rights reservedServerless and Highly Performant Search Engine designed for Pr