aulx
aulx copied to clipboard
Sophisticated score system using user selection frequencies
So far, the idea is to compute the final score of each suggestion based on a weighted sum.
For instance, on JS:
w1·(keyword score) + w2·(static analysis score)
+ w3·(dynamic lookup score) + w4·(user frequency score)
with w1 < w2 < w3 < w4
.
The user frequency score would be attached to display names, instead of accurately scoped properties, both to prevent wasting memory and to avoid resetting that information when the cursor changes scope.
I am still not sure on integrating the frequency score in the overall score to decide the order of popup items as compared to just use the frequency score to decide the highest weighted entry and highlight it when the popup is open.
because for a ML system to work properly , we would need some historic trailing data to decide the w4