Azure Cognitive Search AMA: Vector search, Azure OpenAI Service, generative apps, plugins & more

Hi Glenn. Vector search and Semantic search are somewhat orthogonal but aim to serve the same function – improve the quality of the search results. In this context, Vector search is used to improve recall – the number of relevant results returned by a search query. Semantic search improves ranking by bringing up the most relevant documents to the top of the results list. Vector search and Semantic search can be used independently but work best when used together. Note, Semantic search today works only for search requests that include a text query. The effectiveness of each method will depend on your scenario. For example, Vector search is only as good as the model you use to vectorize the data and you should refer to the model benchmarks to understand its strengths and weaknesses. Semantic search leverages Bing models trained on the web corpus, so it’s very effective for data that’s not domain-specific, but it’s best to test yourself on your data (Semantic search is free for the first 1000 requests). Please let me know if this answers your question.