I have performed a query using Pinecone which has the following results:
0.806332707 - It’s never been easier to add the tawk.to widget on your Cloudflare website: Help Center | tawk.to | Adding a tawk.to widget to your website with Cloudflare
0.804840088 - We have a feature called Sneak Peek which allows agents to read a customer’s message while they are typing in the chat box.
So to verify I manually computed the cosine similarity between query embedding and return values from top_k using
np.dot(vector_a, vector_b) / (norm(vector_a) * norm(vector_b)) and its same.
But when I take the same content and compute embedding and similarity, I see differences in 1 entry. Why is that?