Looking up how to compare vectors often leads to the use of cosine similarity, a popular method for determining similarity between embeddings. The process involves calculating the cosine of the angle between two vectors, with 1 indicating identical directions, 0 being orthogonal, and -1 meaning opposites. This method is ideal for comparing high-dimensional embeddings as it focuses on direction rather than magnitude. Surprisingly, the cosineSimilarity function in JavaScript does not utilize Math.cos() due to computational efficiency. While cosine similarity is widely used, alternatives exist and will be explored in future posts. The power of embeddings lies in their multidimensionality, showcasing the versatility and simplicity of cosine similarity.
https://tomhazledine.com/cosine-similarity/