What a base paper is, and how to choose one

What does “base paper” mean in an Indian PhD?

The idea comes from engineering project culture. Final-year BE and ME students are often given a “base paper” to implement, and the habit carries into the PhD. A supervisor or the scholar picks a recent paper that solves a problem close to the thesis topic. The scholar implements it, confirms the results, and then improves on it: a better algorithm, a new constraint handled, a wider test.

It is a working term used by supervisors and committees, not a requirement you’ll find in the UGC PhD regulations. Used well, it gives a first-year scholar something concrete to do while the literature review takes shape, and it produces a working baseline for every later experiment.

Used badly, it narrows the whole thesis to one paper. Examiners notice when every chapter is “the method of [12], modified”. A base paper is a starting point, not the boundary of your work. Many scholars end up with one base paper per objective, and a literature review that places each within the wider field.

How do you choose a good base paper?

Collect three to five candidates from your reading and judge them against the same criteria.

Recent. Usually within the last five years. An older paper is fine if it is a benchmark that recent work still compares against, but then your comparison must include the newer methods too.

From a genuine indexed journal. Check the journal yourself in Scopus Sources or Clarivate’s Master Journal List; our journal checker takes you to the official records. A paper from a predatory or discontinued journal is a weak foundation, and examiners will ask why you trusted it. Our predatory journal guide lists the warning signs.

Reproducible. Read the method section as if you had to code it tomorrow. Are the equations complete? Are parameter values, data splits and evaluation measures given? Short conference papers often leave these out.

Data and code available. A public dataset, a repository link or a supplementary file saves months. Some journals and conferences award badges for this; the ACM, for example, uses an “Artifacts Available” badge when the materials are placed in a permanent public archive.

Feasible for you. A method that needs eight GPUs, a licensed simulator your department lacks, or a field site in another country is the wrong base paper, however good it is.

Read each candidate critically, as our guide to critiquing a research article describes. The weaknesses you find are where your contribution will come from.

Why reproduce the base paper first, and what if the results don’t match?

Reproduce before you extend, for three reasons. You learn the method in a way reading never teaches. You get baseline numbers on your own data split, hardware and software, which is the only fair comparison for your improved method. And you find out early if something in the paper is wrong.

The ACM’s artifact review policy draws a useful distinction. Reproducibility means an independent group gets the same result using the original authors’ own artifacts (their code and data). Replicability means getting it with artifacts developed independently. If the authors share code, you are reproducing; if you write the code yourself from the paper, you are replicating, which is harder and more likely to show small differences.

Set a clear target (“Table 4, accuracy on dataset X”) and an acceptable tolerance, and record your environment: library versions, random seeds, hardware. Give it a time box, say six to eight weeks, agreed with your supervisor.

When the results don’t reproduce

1. Check your own implementation first, line by line against the paper: preprocessing, data split, parameter values, evaluation formula.
2. Run several seeds. A single run can differ from a reported average by more than you expect.
3. Look for supplementary material, errata and the authors’ code.
4. Write to the corresponding author with specific, polite questions (“Was the data normalised before or after the split?”). Many authors reply.
5. If the gap remains, report it honestly. Use your reproduced numbers as the baseline, state the difference from the published figures, and explain what you tried. Never put the paper’s published numbers in your results table as if you had obtained them.

A documented failure to reproduce is a finding, not a disaster. If the gap is large, discuss with your supervisor whether a different base paper would serve the thesis better.

How do you show a contribution beyond the base paper?

The most common examiner complaint about base-paper theses is that the contribution is a tweak. The difference is usually in the reason, not the size of the change.

Start from a gap, not from a modification. State what the base paper cannot do, why that matters, and what your work does about it. An illustrative example, invented for this guide: “The intrusion detection method of [7] assumes that the traffic classes in training data are balanced. Real IoT traffic is heavily imbalanced, and in our reproduction the method’s recall on the rarest attack class fell to 41%. This thesis develops a cost-sensitive variant and tests it on three imbalanced datasets.” That names a limitation, shows evidence for it from the reproduction, and says what follows. Our research gap statement guide gives a pattern for writing this.

Then make the evaluation fair. Compare against the base paper and against other recent methods, use more than one dataset, report the average and spread over several runs, and include an ablation study showing which part of your change produced the gain.

How do you cite the base paper and avoid similarity problems?

Cite the base paper every time you use something from it: the problem formulation, equations, algorithm steps, datasets, figures and results. In results tables, label columns clearly, for example “reported in [7]” and “reproduced (this work)”.

The similarity risk sits in the method chapter. Scholars who have spent months inside one paper tend to describe its method in its own words, and similarity software flags copied sentences whether or not they are cited. Write the description fresh from your understanding, keep standard equations as equations, and quote directly only a sentence you truly can’t improve on. Our paraphrasing and quoting guide shows how, and the plagiarism limits guide sets out the UGC 2018 similarity levels and how universities apply them.

Figures need care too. Redraw a diagram in your own style and caption it “adapted from [7]”. Reproducing a published figure unchanged may need the publisher’s permission, so check the journal’s permissions page. If you use the authors’ code, follow its licence and cite the repository as well as the paper.

What do examiners ask about the base paper at the viva?

Expect some version of each of these, and prepare a short, honest answer:

• Why did you choose this paper and not another?
• Did you reproduce its results? How close did you get, and how do you explain the difference?
• In one sentence, what does your work do that the base paper does not?
• Why didn’t you compare with [a newer method]?
• Is your improvement larger than the variation between runs? Did you test that?
• Does your method keep any of the base paper’s assumptions? When would it fail?
• What would the base paper’s authors say about your criticism of their work?

The third question matters most. If you can’t answer it in a sentence without mentioning a parameter, work on the contribution before you work on the slides. Our PhD viva questions guide covers the rest of the viva.

Checklist: how do you evaluate a candidate base paper?

Score each candidate against this list before you take it to your supervisor. A paper that fails the first three should be dropped whatever its other merits.

• Published within roughly the last five years, or a classic that recent papers still compare against.
• In a journal or proceedings you have checked yourself against Scopus Sources or Clarivate’s Master Journal List, not taken from the journal’s own website.
• No correction, expression of concern or retraction notice on the publisher’s page.
• The method is described fully: equations, algorithm steps, parameter values, data preprocessing and evaluation measures.
• The data are public or obtainable, and the code is available or the method is clear enough to write it yourself.
• You can run it with the hardware, software licences and lab equipment your department actually has.
• It states limitations, or you can see ones it doesn’t state, that connect to a problem you care about.
• Your supervisor agrees it is a sound starting point and not already worked to death by other scholars in your department.
• You can explain every step of its method in your own words.