Currently, increasing (decreasing) the confidence score of a word by 1 decreases (increases) its probability weight in the flashcard system by 1 (subject to a fixed shift such that the lowest weight is 1).
Subjectively, my feeling is that this doesn't reduce the probability of words quickly enough as you boost the confidence.
I'd like to experiment with a system where there is some decay factor R, and a word with confidence C has a probability weight R^(-C). This way, an extra 1 confidence point causes the relative weight to reduce by a factor of R.
The weights could be scaled by R^M, where M is the maximum confidence score in the word set. This would mean the minimum weight would still be 1.
Currently, increasing (decreasing) the confidence score of a word by 1 decreases (increases) its probability weight in the flashcard system by 1 (subject to a fixed shift such that the lowest weight is 1).
Subjectively, my feeling is that this doesn't reduce the probability of words quickly enough as you boost the confidence.
I'd like to experiment with a system where there is some decay factor R, and a word with confidence C has a probability weight R^(-C). This way, an extra 1 confidence point causes the relative weight to reduce by a factor of R.
The weights could be scaled by R^M, where M is the maximum confidence score in the word set. This would mean the minimum weight would still be 1.
I'll start with R=2 and see how things look.