Amazon Elasticsearch Service now supports Learning to Rank to improve search relevancy ranking

Posted on: Jul 27, 2020

Amazon Elasticsearch Service now supports the open source Learning to Rank plugin that lets you use machine learning technologies to improve the ranking of the top results returned from a baseline relevance query. With Learning to Rank (LTR) support, you can tune the search relevancy and re-rank your Elasticsearch query search results in information retrieval, personalization, sentiment analysis and recommendation systems.  

Elasticsearch, by default, uses BM-25 (BM stands for Best Matching) for search, which relies on the frequency of query terms appearing in each document, to return the most relevant documents. LTR is applied on top of these results to re-score these documents’ ranking based on recency, popularity ratings, personalization, and other behaviors. LTR leverages Elasticsearch queries as feature inputs to the models that are generated and trained using the XGboost and Ranklib libraries in the plugin. These models are deployed in Elasticsearch and applied at search time, making machine learning based ranking accessible for enterprise search.

The LTR plugin is supported on all domains running Elasticsearch 7.7. To learn more, see the documentation.  

Support for Learning to Rank in Amazon Elasticsearch Service is now available in 24 regions globally: US East (N. Virginia, Ohio), US West (Oregon, N. California), AWS GovCloud (US-Gov-East, US-Gov-West), Canada (Central), South America (Sao Paulo), EU (Ireland, London, Frankfurt, Paris, Stockholm, Milan), Asia Pacific (Singapore, Sydney, Tokyo, Seoul, Mumbai, Hong Kong), Middle East (Bahrain), China (Beijing – operated by Sinnet, Ningxia – operated by NWCD), Africa (Cape Town). Please refer to the AWS Region Table for more information about Amazon Elasticsearch Service availability.