The Semantic Search Feature enables the editor to search for content items by their meaning. Instead of matching keywords, the search finds content items that are semantically similar to the search query. For this the content is represented as vectors in a high-dimensional space, called embeddings. These embeddings are generated using machine learning models that capture the meaning of the content. The Content Feeder and the Studio Server connect to a managed Embedding Service provided by Amazon Bedrock in the default configuration.

Please note that you need to license the Semantic Search Feature to use it in your environment. Please contact the CoreMedia support for more information.

Caution

Please note that Amazon Bedrock may incur additional costs depending on your usage.

Figure 4.2. Semantic Search Architecture

The Content Feeder with semantic search enabled sends text, images and textual download data to the Embedding Service to retrieve embedding vectors for each content item. These embeddings are stored in the search index alongside the other indexed content data.

The Studio Server receives the search request from Studio. When a user submits a query, the Studio Server sends the query term to the Embedding Service which now generates an embedding for the query. This embedding is used to perform a vector similarity search against the indexed content embeddings to find the most relevant results. This search is performed by Solr.

This architecture enables semantic search capabilities, allowing users to find content based on meaning and context rather than just keyword matching.

The blueprint workspace includes example configurations for both embedding service integrations: