I'm working on a project where I need to vectorize a list of JSON objects representing products. The challenge is that AI Search only allows vectorization on one field at a time. To work around this, I've been concatenating all fields into a single string formatted like this: : . : . While this method has worked for many products, I'm running into issues with certain queries, particularly those involving product IDs. I'm looking for advice on two things: first, is there a more automated way to improve search quality? And second, is my current approach of concatenating everything into one string the best option, or could it create problems down the line?
3 Answers
Have you tried directly asking CoPilot for suggestions? While I know your current implementation is based on it, there might be features or optimizations you aren't aware of. It sounds like they could offer you some better insights on how to manage this.
You might want to consider treating each product as a separate document instead of merging everything into one. This allows you to manage the properties independently. Also, try to make sure your concatenated string reads naturally, almost like a sentence. Depending on your AI Search capabilities, you could explore setting up different indices for each field and running searches across all of them simultaneously. Just a thought, though—why would you need to vectorize a product ID specifically?
If you're concatenating keys and values, it’s important that they make logical sense together. It sounds like your current method works, but it might be worth looking into indexing the different fields separately if the search API allows for that. If you're forced to concatenate because of how you're pulling data into Copilot Studio, then keep at it, but be aware of potential complications.
Related Questions
xAI Grok Token Calculator
DeepSeek Token Calculator
Google Gemini Token Calculator
Meta LLaMA Token Calculator
OpenAI Token Calculator