AI for Education

Hint Generation and Evaluation

Design, generate, and mathematically evaluate step-by-step hints for open questions.Stop guessing and start measuring hint quality.

01

Design

Start by inserting a factual question.

02

Generate

Use LLMs to create multiple hint sequences that guide students step-by-step.

03

Evaluate

Analyze metrics like Leakage and Convergence to pick the perfect hints.

Evaluation Metrics

Convergence
Range: 0.0 - 1.0

Measures how strongly the hint steers the learner towards the specific target solution.

Leakage
Range: 0.0 - 1.0

Detects if the hint inadvertently gives away the answer too early in the sequence.

Familiarity
Range: 0.0 - 1.0

Estimates if the concepts used are appropriate for the target student's knowledge level.

Relevance
Range: 0.0 - 1.0

Ensures the hint is contextually related to the solution path without drifting.

Readability
Range: 0 / 1 / 2

Classifies linguistic complexity: Easy (0), Intermediate (1), or Difficult (2).

Example: "Capital of Australia"

Question

Which city is the capital of Australia?

Generated Hints

Hint 1

It is not the largest city.

Hint 2

It was chosen as a compromise and is inland.

Hint 3

The name starts with 'C' and houses Parliament.

HintConv.LeakRead.
Hint 10.450.960.94
Hint 20.700.880.90
Hint 30.920.620.86
* Note how Hint 3 has high convergence but lower leakage avoidance (it gives away too much).