groonga
groonga copied to clipboard
ii: use TF-IDF like score for similar search
Old Groonga uses int for score data type. Recent Groonga uses double for score data type.
Now, we can use floating point number for score.
How about using TF-IDF like score for similar search? This change doesn't change matched records order. It just changes score values to smaller values (around 1.0) from larger values. For example, 419432 -> 1.11 and 209716 -> 0.55. Smaller values will be useful for combining score computed by other search methods.