Reading a Systems or Machine-Learning Paper · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.
En anglais

Reading the evaluation first

Computing · Software Engineering · ages 22-23
Name ______________________   Date ____________

  1. A system returns more documents. What usually happens to the two scores?

    • Recall rises, precision usually sinks
    • Both scores always rise together
    • Both scores freeze whatever is returned

  2. What is your first job when opening an unfamiliar paper?

    • Memorizing the method details
    • Say the claim in one sentence
    • Counting the references

  3. A win on one easy data set proves the model is better in general.

    Circle one:   True   False

  4. Why read the evaluation before the method?

    • Methods are never worth reading
    • To see what the results actually show
    • Evaluations are shorter than methods

  5. A paper quotes only precision for its retriever. What do you ask next?

    • The font of the results table
    • The authors' favorite baseline
    • Recall on the same setup

  6. A paper beats a weak baseline and declares victory. What is the sharpest question?

    • Which strong baseline was skipped
    • How many pages the related work spans
    • Whether the title could be punchier

  7. A paper pairs thin numbers with vivid user stories. How do you read it?

    • Let the stories carry the claim, since numbers bore readers
    • Test numbers and stories separately
    • Average the stories into the tables

  8. A result rests on users typing normally, but half typed with gloves on. What changed?

    • Nothing, since data is data
    • The load-bearing assumption failed
    • The method section grows by one paragraph
LightMySky · lightmysky.comW1-mt_4LsuT0TdxT-s1

Answer key

For grown-ups. Fold this page away before handing over the rest.

Reading the evaluation first W1-mt_4LsuT0TdxT-s1

  1. Recall rises, precision usually sinks · More returns catch more relevant items plus more irrelevant ones.
  2. Say the claim in one sentence · Claim first, then the evidence offered for it.
  3. False · The evidence covers that set only; the broad claim needs broader proof.
  4. To see what the results actually show · Demonstrated facts first, then how the machinery might explain them.
  5. Recall on the same setup · One side of the tradeoff hides the other; demand both.
  6. Which strong baseline was skipped · Victory over the weak says little until the strong are faced.
  7. Test numbers and stories separately · Each tool must earn its own keep against the claim.
  8. The load-bearing assumption failed · Assumptions carry results; broken carriers drop them.
Worksheet · LightMySky