I'm working on computing the cosine similarity between two vectors in my project. While I understand that the standard cosine similarity ranges from $-1$ to $1$, I'm interested in transforming the similarity scores into a range of $[0, 1]$. I have two main approaches in mind, but I'm unsure about the best way to proceed.
Scaling Approach: simply scaling the cosine similarity values to the $[0, 1]$ range. Would this approach be valid and reliable? Are there any potential drawbacks or issues I should be aware of when applying this method?
Angular Similarity: which measures the angle between the vectors directly. Would using angular similarity inherently provide similarity scores in the $[0, 1]$ range, and is it a better alternative to the scaling approach?