Using CMS to Load Batch Analytics as ML Features
A common use pattern to use internal analytics pipelines to improve ML performance.
The Item and User CMS can be used to load outputs of regular (e.g., daily) batch analytics pipelines that generate aggregate statistics and model inferences per item and per user. Many companies use these pipelines to generate reports and power internal ML systems. For example: 356dayPurchaseCount, itemQualityScore, userLastSeen.
Any such statistics loaded to any CMS system are automatically added to all models using Promoted's autoML systems. There is no special configuration necessary to use this integration pattern. Any reasonable JSON format is supported, see CMS Feature Formats. Use separate CMS sourcefor such daily analytics pipelines to keep the records isolated from other item attribute records using the CMS pattern of one-source-per-writer.
Updated about 2 months ago