EC-17.2 Finding the Literature, and Reading a Paper

What this is and why it exists

A student who cannot find prior work will spend a capstone rediscovering something published in 2009.

This topic teaches the search, the triage and the reading order. It is deliberately practical rather than a survey of databases. The skill is in narrowing a query, and in deciding quickly whether a paper is worth an hour.

Most papers you find are worth five minutes. Knowing which are the exceptions is the point.

The vocabulary

• Prior art — what already exists that addresses your problem.
• Peer review — evaluation by other researchers before publication.
• Preprint — a paper made public before peer review.
• Reference list — what a paper built on.
• Citation — a later paper that refers to this one.
• Baseline — the comparison a result is measured against.
• Reference manager — a tool that stores citations in a reusable form.

The mental model

Searching is a vocabulary problem before it is a database problem. The terms an author would use are rarely the terms you would.

The efficient route is indirect. Find one relevant paper by any means, then mine its title, keywords and reference list for the vocabulary the field actually uses. Search again with those words. Two rounds of that beats an hour of guessing better search terms.

Once you have one good paper, follow it in both directions. Its reference list leads backwards to what it built on. The papers citing it lead forwards to what came after.

Two hops in each direction from one well-chosen paper usually covers a small field completely. It also shows you which works everybody cites, which is a fast way to find the ones worth reading properly.

Now triage, and the distinction that matters most. Peer review means other researchers examined the work before it was published.

That raises the floor. It does not guarantee correctness, and reviewers routinely miss things. A preprint has had none of that scrutiny, which does not make it wrong, only unchecked. A vendor white paper is advertising, however many graphs it contains. Read it as an assertion by an interested party.

Read a paper in three passes and stop as soon as you have what you need.

First pass: title, abstract, conclusion. Decide whether to continue. Most papers stop here and that is the correct outcome.

Second pass: figures and results. What did they measure, and how does it compare with what you need? For most of the remaining papers this is enough.

Third pass: the method, in detail, only if you intend to use it or to check it. This is the expensive pass and it should be reserved for the two or three papers that matter.

Reading the results section critically means asking three things. What was measured? Under what conditions? Compared against what?

A result with no baseline tells you nothing, because there is nothing to be better or worse than. A result reported only for the favourable configuration is a finding about the authors rather than about the subject. It is recognisable by the absence of configurations you expected to see.

Finally, keeping references usable. Full citation, where you found it, and one sentence on why it mattered to you.

That third item is what makes a reference list useful a year later. Reconstructing a half-remembered paper from a phrase costs an afternoon, and recording it properly cost thirty seconds. A reference manager does the formatting; the sentence is yours to write.

What you should now be able to explain or do

• Find the vocabulary a field uses and search with it rather than with your own terms.
• Follow references backwards and citations forwards from a single good paper.
• Distinguish peer-reviewed work, a preprint and a vendor document.
• Read a paper in three passes and stop at the right one.
• Read a results section critically, checking measurement, conditions and baseline.
• Record a reference so that it is usable a year later.

Check yourself