aws AWS Big Data Blog ·

Delivery Hero migrates search to Amazon OpenSearch Service using radial search

blogdataawsengineer
announcement

This post details how Delivery Hero transitioned their semantic search infrastructure to Amazon OpenSearch Service, replacing a self-managed Kubernetes-based system. The migration enabled them to leverage radial vector search combined with lexical search for improved relevance, scalability, and cost-effectiveness. The new architecture utilizes OpenSearch Ingestion for data syncing and offers a flexible approach to hybrid search that balances recall and precision.

  • Hybrid search with radial vector and lexical retrieval
  • Radial search for improved relevance and variable result sets
  • Delivery Hero's search infrastructure modernization
  • Ingestion pipeline using Amazon OpenSearch Ingestion
  • ID resolution and inventory filtering within a single call
Features (2)
  • Hybrid search with radial vector and lexical retrieval

    The inference pipeline employs a hybrid search strategy, encoding user queries into embeddings and running parallel radial k-NN search against product embeddings and lexical BM25 search against the product catalog.

  • Radial search for improved relevance and variable result sets

    Radial search, using the min_score parameter, was chosen over traditional k-NN to retrieve all semantically relevant products above a similarity threshold, offering better quality results and a variable result set size adapted to query specificity.

Enhancements (1)
  • ID resolution and inventory filtering within a single call

    The system includes a resolution step that maps product parent IDs from the k-NN index to individual product IDs via a secondary index, incorporating near real-time inventory updates within a single retrieval call.

Notes (2)
  • Delivery Hero's search infrastructure modernization

    Delivery Hero migrated their entire search infrastructure from self-managed Elasticsearch on GKE to Amazon OpenSearch Service 3.x, consolidating their semantic search service into the managed platform.

  • Ingestion pipeline using Amazon OpenSearch Ingestion

    An ingestion pipeline was established using Amazon OpenSearch Ingestion (OSIS) to sync product embedding data from Amazon S3 to OpenSearch Service, refreshing the k-NN index daily for optimized performance.

Read the original announcement →

https://aws.amazon.com/blogs/big-data/transforming-search-at-delivery-hero-a-migration-journey-to-opensearch-service-with-radial-search/

Related releases