F1 · Numbers That Learn
Agreement Without Volume
Score two vectors on direction alone, and say when that beats a dot product.
The dot product has a flaw you can see. Take a review that only half matches your taste, and make every number ten times bigger. Its taste did not change, but its score did. Now it beats a quieter review that matches you perfectly. The fix is called cosine similarity. It scores the direction only and throws the lengths away.