How can I create a custom search function in Cognitive search?

Question

Hi,

I want to combine keyword based and embedding based search, and found out that OpenAI Embeddings can be used as a custom skillset in Cognitive search to enrich the data set. My idea was to find a way to both use cognitive search word search and similarity search using OpenAI embeddings, and create a similarity score based on a weighted average.

I followed this repo to set up data enrichment of the indexes using OpenAI Embeddings: https://github.com/Azure-Samples/azure-search-power-skills/tree/main/Vector/EmbeddingGenerator, however, there is no mention of how to use the embedding vectors in a similarity search. I have not found a way to add a custom similarity search function in order to actually use the embedding vectors to perform a similarity search. Is there anything I have overlooked?

Answer 1

If you are using Python check out the link below. There are examples of using both vector and text combined. Note the setting of search_text in the examples. If set to "" and then using the vector parameter you will do a Vector only search. If you set it to your query you can do hybrid.

https://github.com/Azure/cognitive-search-vector-pr/blob/main/demo-python/code/azure-search-vector-python-sample.ipynb