What Goes in the Abstract of a Computer Science Thesis? (2026)

A computer science abstract is read by more people than any other 300 words you will write, and it is judged by a different standard than an abstract in the humanities or social sciences: a CS reader wants to know, in order, what problem you solved, how, and how well it worked — measured against something. Generic “how to write an abstract” advice does not tell you what a systems, algorithms or machine learning examiner actually expects to see in that order, or which keyword system your department wants you to use alongside it.

1. The Four Moves of a CS Abstract

A strong computer science abstract makes four moves, almost always in this order, and almost always without headers separating them:

• Problem — the specific gap or limitation, stated concretely (a task current systems handle poorly, an inefficiency, an unsolved case), not a broad claim that “technology is important.”
• Approach — what you built, trained, proved or evaluated: the system, architecture, algorithm or protocol, named precisely enough that a reader in the sub-field recognises the design family.
• Result — the number. An accuracy, F1 score, latency reduction, throughput gain, proof of correctness, or head-to-head comparison against a named baseline. A CS abstract without a quantitative or formally stated result reads as unfinished, even if the thesis itself has one deep in the results chapter.
• Significance — one closing sentence on why the result matters beyond the specific benchmark: what it enables, what it corrects, or what open question it resolves.

Examiners reading dozens of abstracts a season learn to skim for these four moves specifically — a reviewer or supervisor who cannot locate the problem, the approach, the number and the significance within the first read is likely to ask you to redraft before the thesis goes any further, regardless of how strong the underlying work is.

2. A Worked Example (Illustrative)

Below is an illustrative abstract following the four-move structure for a hypothetical machine learning thesis, with invented but clearly labelled figures:

Notice the abstract names the dataset, states the metric, gives a specific comparison point, and never says “in recent years” or “with the rapid growth of the internet” — the kind of narrative throat-clearing that eats word count without adding information a CS reader needs.

3. What a CS Abstract Skips

Conventions that are standard in social-science or humanities abstracts are usually absent from a strong CS abstract:

• Broad scene-setting — “as technology continues to evolve” openings waste the first sentence a reader will actually read.
• Literature-review summary — a CS abstract does not summarise prior work in general terms; it names the specific baseline or gap your result is measured against.
• Methodology philosophy — no positivist/interpretivist framing; a CS abstract states what was built and evaluated, not an epistemological stance.
• Hedged findings without numbers — “the results suggest a possible improvement” without a figure reads as evasive in a field where a number is expected.

4. Keywords: ACM CCS Concepts vs IEEE Terms

Many computer science departments expect a structured keyword or concept list immediately after the abstract, separate from index terms in the body text. Two systems dominate depending on your target publication style:

• ACM CCS Concepts — the ACM Computing Classification System’s hierarchical concept taxonomy (for example, “Computing methodologies → Machine learning → Supervised learning”), used in ACM-format theses and papers, typically listed with an assigned relevance level.
• IEEE index terms — a flatter list of free-text or IEEE Thesaurus-drawn keywords, used in IEEE-format work, listed directly beneath the abstract without the hierarchical structure.

Check your department’s required citation and formatting style before assuming either convention — a thesis submitted in IEEE format with ACM-style CCS Concepts (or vice versa) is a formatting error examiners notice immediately. For the full citation-style decision beyond just the abstract, see our IEEE citation style guide.

5. How the Emphasis Shifts by Sub-field

The four-move skeleton is constant, but what makes a credible “result” differs sharply by sub-field, and using the wrong kind of evidence for your sub-field is a common reason a draft abstract reads as unconvincing to a specialist examiner even when the underlying work is sound.

6. The First Paragraph of Your Introduction

The first page of the introduction should not restate the abstract in longer sentences. Its job is different: establish the problem with enough technical detail that a reader outside your immediate sub-field can follow it, cite the two or three works that define the gap you are addressing, and state your specific research questions or contributions as a numbered or bulleted list before the chapter goes any further. A CS introduction that opens with the same problem sentence as the abstract but never sharpens it into a testable question by the end of page one is a common source of examiner confusion about what, precisely, the thesis claims to contribute. For the full chapter-by-chapter structure this leads into, see our guide on how to write a computer science dissertation step by step, and for the wider methodology-chapter conventions, see our guide on writing a research methodology chapter for your thesis. If your thesis is specifically machine-learning-focused, our guide on choosing a methodology for a data science or machine learning dissertation covers the evaluation-protocol details that feed directly into the “result” move of your abstract.

7. How the Abstract Differs by Degree Level

The four-move structure holds across degree levels, but what counts as an acceptable “result” scales with the stakes of the degree:

• Undergraduate final-year project — a working implementation evaluated against a clear, if modest, baseline is usually sufficient; the abstract can state a single headline metric without needing to claim novelty over the published literature.
• Master’s thesis — the result should be benchmarked against at least one existing published approach, not just an internal ablation, and the significance sentence should gesture at a genuine (if narrow) contribution.
• PhD thesis — the abstract typically summarises several linked contributions (often one per publication or chapter) rather than a single experiment, and the significance move usually states what the overall body of work establishes that no single chapter does alone.

8. Mentioning Code and Data Availability

An increasing number of computer science venues and departments expect a one-line reproducibility statement either inside the abstract or immediately beneath it — a note that code, trained models or the evaluation dataset are available at a named public repository, or an explicit statement of why they are not (proprietary data, an industry partnership NDA, patient-level records). Leaving reproducibility entirely unaddressed reads as an oversight in venues where it has become a default expectation, particularly in machine learning sub-fields where unreproducible headline numbers have become a recognised credibility problem. If your thesis cannot release code or data, say so explicitly and briefly rather than leaving the question unanswered.

9. Common Mistakes

• No number in the abstract — stating a method was “evaluated” without saying what the evaluation showed leaves the single most-scanned sentence in the thesis empty of the information it exists to carry.
• Comparing against nothing — a result with no named baseline gives the reader nothing to calibrate the number against; “94% accuracy” means little without knowing what a simple baseline or the best prior published method achieves on the same data.
• Front-loading background instead of the problem — three sentences of history before the actual gap is named pushes the one sentence a skimming reader needs to the middle of the paragraph instead of the start.
• Copy-pasting the abstract as the introduction’s opening paragraph — the introduction should sharpen, not repeat, what the abstract already said, with more technical detail and named prior work.
• Wrong keyword system for the required citation style — ACM CCS Concepts in an IEEE-format thesis, or vice versa, is a formatting inconsistency examiners notice on the very first page.

Frequently Asked Questions