Can't find any post about this so i make one instead. If i grade a character as 4 it will for sure come again in some other word combination. If i then don't change the grading of that character when it comes again, will it be 3 or remain 4? If it is 3 does that mean it will come back earlier? I guess that if a new word is added that has two grade 4 characters and i do not change anything when the words come up the word will be grade 3. In another example, if i grade a word as 4 but not the characters, they will still be grade 3 i guess? Just trying to get the picture here. I recently nuked my account,again, to prepare for an upcoming test. After that some of the characters behaves weird the first time they come back.

A character doesn't have a grade in the system at any point; it just has a review interval. Here's an example. Say you add 我 and grade it as a 4, so it's scheduled for, say, a month.

Two weeks later, you add 我们 and mark 我 in it as a 3. It will increase the review interval by about half of what it would have if you had practiced it on time, so half of a about two months is another month from now instead of two weeks.

Does that make any sense? If you had instead studied 我们 right after studying 我, Skritter would have hardly changed the interval at all, since you'd only get a tiny fraction of the interval increase you'd get if you'd practiced it as scheduled. It definitely doesn't come any sooner.

I have a related question. Do you guys track the differences in the forgetting rate for each item? That is, there could be differences between items as far as how difficult they are to remember after X amount of time.

Let's say you have two items which the user has never seen before, A and B. After a couple of repetitions, the user gets them both right, so you schedule them out for review in a day. After a day the user reviews them and gets A right and B wrong, so you schedule A out for a week and B for immediate review. Then let's say B eventually gets scheduled out for a week, and the user again gets it wrong, etc.

In this case I would say that there seems to be something intrinsically different about B which makes it more difficult to remember. The fact that it might be currently scheduled for a month out doesn't capture that difference, because both A and B might be scheduled for a month from now.

James

Only in two cases. If you never get an item wrong, its intervals grow much faster for the first few reviews, to help get easy items out of the way.

And if you get an item wrong more than half of the time (for more than eight reviews), its intervals will grow more slowly.

Later I want to measure and integrate per-item toughness (across all users) into the scheduling, to predict the sort of per-item difficulty that you're talking about.

I get the impression that a big part of the difficulty for certain items is interference. It would be helpful to identify groups of items that interfere with one another and split them up. And maybe there are some items that really are fundamentally more difficult - it would be useful to be able to identify them to pull them out, cram them, study their compounds, etc.

I have an idea about how to estimate difficulty using a sequential Bayesian process if you're interested in that kind of thing.

James

Interference was a big problem for me. Now if I'm confusing two or more characters, I make a note of them in the custom definitions field. A small hint I guess, but big benefits. It mutually reinforces the characters rather than confusing the hell out of me.

Yeah, when I notice that an item is hard, I also do something like that. I was thinking more in terms of the system identifying difficult items that we're not aware are difficult. I recently took a two-month hiatus and as a result many of the items I ran into hadn't come up in a long time, and as usual some of the long-buried items were mysteriously easy (I don't think I've ever gotten 空 wrong, for some strange reason) while others were mysteriously hard. I don't imagine that I can actually identify them myself, though; they might be hard without me noticing, right?

James