LogoGADvance

How BKT works

How Bayesian Knowledge Tracing estimates your mastery after each answer and chooses question difficulty.

On this page

The idea in one minute#

Bayesian Knowledge Tracing treats "has this learner mastered the topic?" as a probability between 0 and 1. Every time you answer a question on a topic, GADvance updates that probability. A correct answer usually raises it and an incorrect answer usually lowers it, but not by a fixed amount. BKT accounts for lucky guesses and careless slips, so one answer never decides everything.

One BKT updateYour current mastery estimate and your answer are combined using the guess and slip probabilities, then the learning transition probability is applied to produce the new mastery estimate, which decides the next question difficulty.Current masterye.g. 0.20Your answercorrect / incorrectWeigh evidenceguess + slipLearning+ transitionNew masterysets next difficulty
Each answer is weighed against the guess and slip probabilities, then a small learning step is applied. The result sets the difficulty of the next question.

The four parameters#

Initial Knowledge

P(L₀) = 0.20

How likely you are to know a topic before answering anything about it.

Transition Probability

P(T) = 0.10

The chance you learn the topic from each opportunity to practice it.

Guess Probability

P(G) = 0.25

The chance of answering correctly without knowing the topic.

Slip Probability

P(S) = 0.10

The chance of answering incorrectly even though you know the topic.

These are the values GADvance currently uses for every topic.

How an answer updates mastery#

Each update has two steps. With current mastery L:

Step 1: weigh the evidence

If correct: L' = L(1 − S) ÷ [L(1 − S) + (1 − L)G]

If incorrect: L' = L·S ÷ [L·S + (1 − L)(1 − G)]

Step 2: allow for learning

New mastery = L' + (1 − L') × T

The result is kept between 0.01 and 0.99 and rounded to three decimal places.

Worked example#

A learner starts a topic at the initial knowledge value of 0.20 and answers five questions: correct, correct, incorrect, correct, correct. These numbers are computed with the same formula GADvance uses.

Mastery estimate after each answerStarting at the initial knowledge value of 0.2: start 0.2, correct 0.526, correct 0.82, incorrect 0.44, correct 0.765, correct 0.929.Analyze / EvaluateApplyRemember / Understand0.000.400.701.000.20Start0.53Q1 ✓0.82Q2 ✓0.44Q3 ✗0.77Q4 ✓0.93Q5 ✓
Mastery after each answer. Shaded bands show which question difficulty the learner would be served next.
QuestionAnswerMastery after
Q1Correct0.200 → 0.526
Q2Correct0.526 → 0.820
Q3Incorrect0.820 → 0.440
Q4Correct0.440 → 0.765
Q5Correct0.765 → 0.929

Notice that the incorrect answer on Q3 lowers mastery a lot but does not reset it, and two more correct answers bring it back up.

Choosing question difficulty#

Checkpoint questions are tagged with a Bloom's Taxonomy level. Your current mastery decides which level you see next:

Building

Mastery below 0.40

Remember / Understand

Developing

Mastery 0.40 to below 0.70

Apply

Proficient

Mastery 0.70 and above

Analyze / Evaluate

How long a checkpoint is#

Each adaptive checkpoint has a minimum and maximum number of questions, set by the course. GADvance picks a length between them from your average mastery across the topics the checkpoint covers:

  • Average mastery of 0.50 or lower: the maximum number of questions, for more practice.
  • Average mastery of 0.90 or higher: the minimum number of questions.
  • In between: proportionally fewer questions as your mastery rises.

Mastery is not your score#

Two different numbers

Your score is the percentage of points you earned in one attempt, and it decides whether you pass. Your mastery estimate is a probability that GADvance keeps updating across all your answers on a topic. It decides question difficulty and review suggestions, not whether you pass.