Knowt Implementation of Algorithm

Flow:

• User goes to the note editor screen and clicks “Study Button”

• If they haven’t made flashcards → create flashcards and show them on the note

• They can then add more flashcards or continue the study process

• If they have made flashcards → let user pick how they want to study

• Study Flashcards

• When do you want to know the material?

• Ask users to place their answers into one of the boxes and use our algo to show the cards

• Too Soon

• I forgot

• Not sure

• Too easy

• Keep track of terms studied

• Track progress independently from quizzes

• Show progress bar at the top

• Learn Concepts

• How well do you know the notes already?

• When do you want to know the material?

• Quiz:

• Instant Feedback only

• Create a large quiz and have the user go through the questions

• Spaced out with their choices

• Give “breaks” every 10 question to recap information (maybe show this in the UI)

• Ask users to place their answers into one of the boxes and use our algo to show the questions

• Too Soon

• I forgot

• Not sure

• Too easy

• Bad question box? (Only for quizzes)

• Track progress independently from flashcards

• Show progress bar at the top

• You should be able to resume either from where you left off

• How do we handle switching between the 2?

• Show 2 buttons?

• Let them switch in the settings if they’ve started

• Just ask them at the end

Algorithm:

• We use the same three phases: Learning, Exponential, Relearning

• We adjust the values of each variable based on the following factors:

• The date of their exam

• Intervals should change based on the time we have

• How well they know the note

• Should adjust the quiz difficulty (nothing we can do for flashcards)

• Quiz

• Should follow showing the 4 or 5 buttons

• Buttons should override the user input

• Got it wrong → Clicking easy overrides (mark it as easy)

• Got it right → Clicking hard overrides (mark it as hard)

• Look into defining or adjusting the algo based on incorrect/correct and response to the buttons

• Repeat questions as we are repeating flashcards

• Notifications

• Notify the user to come back (nudge) and continue studying

• See how this adjusts based on the timeline that the user gives us

• 2 days study period → multiple notifications during the day (find a good balance)

• 7 day period → one notification per day

• Questions

• How do we handle the mix of flashcards and quiz questions

• Value is assigned to each term

• Proposed → Make them independent for quiz/flashcards[a]

Anki’s System

Video: https://www.youtube.com/watch?v=cUW10l4ZZjg&ab_channel=RevisingRubies

• Follows the forgetting curve

• Learning starts off steep so you are forgetting a lot constantly

• Curve starts to get shallower as you add learn it

• Spaced Repetition helps to get this curve shallower

• Cards have 4 options: Again, Hard, Good, Easy

• The Space Between Seeing Cards (intervals) normally increases by a factor of 2.5

• Day 1, 3, 8, 21, 54, 137

• Intervals are 2, 5, 13, 33, and 83 days (see how 5 = 2*2.5, 13 = 5 *2.5)

• Learning Phase → Cards on Day 1 and 3 are said to be in this phase (New Cards too)

• Follows the intervals for 1 minute and 10 minute

• Again → 1 minute

• Good → 10 minutes

• When you press Good two times in a row (aka after 10 mins) then the card Graduates

• Graduation Phase

• First interval after the learning phase (Day 3 → So between Day 3 and 8) is now set as 1 day and you don’t see that graduated card until 1 day after

• Intervals after that increase by 2.5

• If you press Good you keep seeing that card until you indicate you forgot it

• Lapse Cards

• If you forget a graduated card it resets

• You see it after 10 minutes until it graduates (You say “good”)

• This time once the card is graduated its multiplication factor is 2.3

• Decrease by 0.2 every time you forget

• Each card has its own multiplication factor

• Edge Cases - If the user doesn’t come back to the app after the desired space it adjusts the spacing

• Anki considers the actual time the card was unseen, not just the time it was scheduled for

• Ex: If the card was scheduled for 5 days but you don’t study for a month

• Good →  the next interval goes from 10 days to something like 20 days (Factor of 2.5)

• Hard → Less than last wait (Factor of 1.2)

• Easy → Aggressively long interval (Factor of 3.25)

• Logic: If you got it right after a long time you don’t need to see it that soon

• Types of Flashcards

• Basic: One directional

• Reversed: Two directions (learning vocabulary)

• Cloze Deletion: Fill in the blank

• Overlapping Cloze: Not free

• Basic Image Card: Images with one direction

• Image Occlusion: diagram labelling (an add-on)

Anki-Sm2 Spaced Repetition Algorithm

Algorithm Values: https://www.remnote.com/p/help/document/8g5xHDmGSotpAR9Pc

• Learning Phase:

• Fixed Preset intervals

• 1, 2, 4 days

• Graduate by either of these 2:

• Pressing easy (far right option)

• Or finishing the fixed preset intervals

• Exponential Phase:

• Computed by: Multiplier * (Last Scheduled Interval + Days Late / Hardness Divider)

• Multiplier: Computed from Last Score (Hard = 1.2, Good = Ease, Easy = Ease * Set Bonus) + Ease

• Ease is unique to each card

• Every card starts at 2.3 and  is bounded between 1.3 and 2.5

• Again (-0.2)

• Hard (-0.15)

• Good (No Change)

• Easy (+0.15)

• Last Scheduled Interval: Days in the previous interval

• Days Late: Number of days your card was waiting after its set study time (edge case if the user doesn’t return to study on set date)

• Hardness Divider: Fixed from last score (Hard: 4, Good: 2, Easy: 1)

• Relearning Phase:

• Activates when you press Again while the card is in the Exponential Phase

• Goes back to fixed intervals

• Graduate by either of these 2:

• Pressing Easy (far right option)

• Or finishing the fixed preset intervals

• New interval after graduation is computed by:

• Multiplier * (Last Scheduled Interval + Days Late + Last Exponential Interval * Lapse Interval Multiplier[b])  / Hardness Divider)

Timings & Buttons

1. Every card can be in the “Learning”, “Almost-Graduated” and “Graduated” case. Based on the timing we might have an “Exam next day” schedule.

2. The “Learning” case has the following schedule for delays for buttons: (0%)[c]

1. The “Easy” brings the card in the “Almost Graduated” state

2. All the rest retain card in the same state (Learning)

3. The “Almost-Graduated” case has the following schedule for delays for buttons: (50%)

3. The “Easy” brings the card in the “Graduated” state

4. All the rest reset card to the learning state

5. The “Graduated” case has the following schedule for delays for buttons: (100%)

5. The “Easy” brings the card in the “Graduated” state

6. The “Partially recalled” brings the card in the “Almost-Graduated” state

7. All the rest reset card to the learning state

8. The “Next day exam” case has the following schedule for delays for buttons (Excludes cards in the graduated state):

[a]answered

[b]Can be configured between 0 and 100

[c]Need values for 2 days

[d]Thought this was going to be based on the exam date?

[e]Also how are these calculated given the delays in the Learning state

[f]Thought this was going to be based on the exam date?

[g]Also how are these calculated given the delays in the Learning state

[h]Thought this was going to be based on the exam date?

[i]Also how are these calculated given the delays in the Learning state

[j]Thought this was going to be based on the exam date?

[k]Also how are these calculated given the delays in the Learning state

[l]Based on the exam date? Relation to previous "Almost-Graduated case?"

[m]Based on the exam date? Relation to previous "Almost-Graduated case?"

[n]Based on the exam date? Relation to previous "Almost-Graduated case?"

[o]Based on the exam date? Relation to previous "Almost-Graduated case?"