Event banner
Event details
Vector search is a new feature in Azure Cognitive Search that is currently in public preview. It is designed to provide more advanced search capabilities by using vectorization techniques to represent documents and queries as vectors in a high-dimensional space. This allows for more efficient and accurate matching of similar documents and queries based on their semantic meaning and context. Vector search can be used in conjunction with the RAG pattern to provide more advanced question-answering capabilities and improve the accuracy of search results.
With vector search, you would need to perform vectorization of content externally before passing it into Azure Cognitive Search. This can be done using various techniques, such as word embeddings or deep learning models, depending on the specific requirements of your scenario.
There are updated samples available that demonstrate the power of vector search in Azure Cognitive Search. You can refer to the Azure Cognitive Search documentation and the SDK documentation for code samples and guidance on vector search implementation and management.
Regarding CosmosDB for MongoDB Core offering vector search, it’s important to evaluate the specific requirements and constraints of your scenario to determine which solution is best suited for your needs. Azure Cognitive Search provides a fully managed, scalable, and flexible search service that can handle various types of data and workloads. It also integrates seamlessly with other Azure services, such as Azure OpenAI Service, to provide more advanced natural language processing capabilities. CosmosDB for MongoDB Core provides a fully managed NoSQL database service that can handle various types of data and workloads. It also provides built-in support for MongoDB APIs and features, such as sharding and replication.
liamca-msft
Microsoft
MicrosoftThanks Blair for the great response! I would like to add that Azure Cognitive Search not only offers Vector Search, but also Hybrid Search which leverages scores from traditional text search as well as vectors, which we (and much of the industry research) has found to offer more effective relevance than just Vector Search. In addition, when you then add our Semantic Search (which is a reranking layer), we find this typically offers the most effective relevance, which is incredibly important, especially when build enterprise ChatGPT apps. We are working on a blog post around the effectiveness of this, so please keep your eye out over the next few weeks here.