CSEG3060 · Research Methodology in Computer Science
Clear and concise writing, audience analysis, report types, IMRaD structure, referencing and visual aids
A brilliant algorithm with a badly documented evaluation is, in practice, indistinguishable from an undocumented idea.
Unit IV · Lecture 1 of 3 · 60 minutes · CO4 · B.Tech (CSE) Semester V
Dr. Mohsin Furkh Dar · School of Computer Science, UPES
→ / Space next · ← back · O overview · N speaker notes · F full screen
Learning outcomes · CO4
Apply the principles of clear, concise, precise technical prose — and revise a verbose passage into one
Conduct an audience analysis and calibrate tone, vocabulary and depth to a reader persona
Select a structure — IMRaD, BLUF, modular, pyramid — and justify the choice for a given document
Distinguish the major types of technical report and construct a complete IMRaD report
Apply IEEE, ACM and APA referencing correctly and consistently, with in-text citations
Design tables and figures that are self-sufficient, accessible and governed by the data-ink principle
The premise
Most readers skim before they read. Structure must allow rapid extraction of the conclusion — which is why front-loading and informative headings are not stylistic preferences but functional requirements.
Every claim must be supported, every source traceable, and every figure interpretable without recourse to the text.
Computing has a well-documented reproducibility problem around benchmarks, model evaluations and dataset descriptions. Ambiguity has cascading cost: a misdescribed algorithm yields incorrect implementations; an unclear requirement yields a flawed system; a vague proposal fails to secure funding.
These two assumptions generate every structural, stylistic and citation convention in the rest of this lecture.
Key concepts
| Term | Definition |
|---|---|
| Technical writing | Drafting, revising and presenting specialised information clearly and accurately for a defined audience |
| Audience analysis | Systematic identification of readers’ background, expertise, expectations and informational needs |
| Conciseness | Expressing an idea in the fewest words without sacrificing clarity or completeness |
| Clarity | The reader comprehends the intended meaning on first reading, without ambiguity |
| Cohesion | Grammatical and lexical linkage between sentences — the local glue |
| Coherence | Logical relationship and orderly progression of ideas across the document — the global argument |
| Style guide | A standardised reference (IEEE, ACM, APA) codifying formatting, citation and linguistic convention |
Clear and concise writing
Not brevity at the cost of meaning — the elimination of redundancy, hedging and verbose construction.
Assigns agency, shortens sentences. Passive is legitimate when the actor is unknown, irrelevant, or less important than the action.
“Many researchers have studied this” → “Twenty-seven studies between 2015 and 2024 examined this”.
Mirror the reader’s cognition: known → unknown, general → specific, problem → solution, claim → evidence.
Adjectives and adverbs should describe, not ornament. Excise very, really, quite, basically, essentially unless they carry precise technical meaning.
Principal conclusion first, in each paragraph and each section. This inverted pyramid respects the professional reader’s time and enables scanning.
Professional documents are drafted, revised, peer-reviewed and edited repeatedly. The first draft is never the final version.
Worked example · principles 1–3, 6
“In this section we will be talking about how the new hashing algorithm that we came up with can be used in order to make the process of looking up data faster than what is currently possible using the old algorithm which has been around for many years.”
“We propose a novel hashing algorithm, HashLITE, that reduces average lookup latency by 47% compared to the standard SHA-256-based indexing scheme (Table 2). The improvement stems from a compact 64-bit digest and an in-bucket Bloom filter that prunes false positives before disk access.”
Named the artefact
Quantified the claim
Cited the evidence
Explained the mechanism
Write for the reader
Discipline, level, role, cultural context. An undergraduate, an industry engineer, a funding reviewer and a domain expert are radically different readers.
Neither insult experts with elementary explanation nor bury novices in jargon. Calibrate by layering.
To learn · to apply · to evaluate · to cite. Each changes depth, examples and prominence of conclusions.
Screen or print; linear or scanned; one sitting or many. Web documents demand scannable subheadings.
Priya Sharma — senior software engineer, fintech startup
Background: B.Tech CS, six years in industry
Purpose: deciding whether to adopt a new consensus algorithm in production
Constraints: has 20 minutes; needs actionable recommendations; knows Raft and Paxos, not your CRDT-based variant
The cardinal rule: write for the reader, not for yourself. Most unreadable student writing is writing addressed to its own author.
Structuring technical content
Introduction — context, related work, gap, contribution
Methods — algorithms, systems, setups, frameworks
Results — findings with quantitative detail
and Discussion — interpretation, limitations, positioning
The prevailing template for empirical papers; the basis of the ACM and IEEE templates and of most theses.
Key conclusion at the outset, supporting evidence after. From military and executive communication.
Bottom line: this proposal requests $250,000 over 24 months to develop a privacy-preserving federated learning framework for healthcare analytics.
Support: (1) existing frameworks fail HIPAA compliance; (2) preliminary results show 30% better privacy guarantees; (3) the team has 14 peer-reviewed papers in the domain.
Discrete numbered sections and subsections — technical reports, software documentation, standards. The IEEE conference template prescribes Roman-numeral main headings with alphabetic subheadings.
Group, summarise, support. Thesis first, then grouped supporting ideas, each with its own evidence. Formally a tree T = {r, S₁, S₂, …, Sₙ} where r is the root thesis and each Sᵢ a sub-thesis with its own subtree — a formalism that lets you validate the logic, not just the prose.
Report writing · typology
| Type | Primary purpose | Audience | Distinguishing features |
|---|---|---|---|
| Research / technical report | Document an original investigation | Sponsors, peers, regulators | Full IMRaD; detailed methodology; long reference list |
| Feasibility report | Assess viability of an option | Management, clients | Comparison tables; cost-benefit; recommendations |
| Project report / thesis | Document a sustained effort | Examiners, supervisors | Extended literature review; explicit contributions; oral defence |
| Progress / interim report | Update on milestones | Sponsors, supervisors | Period-bounded; Gantt charts; risk logs |
| Laboratory report | Record an empirical experiment | Instructors, auditors | Hypothesis-driven; numeric data; strict procedural narration |
| White paper | Persuade practitioners to adopt | Industry decision-makers | Marketing tone; business case; architecture diagrams |
| Technical memorandum | Communicate decisions internally | Internal teams | Concise; problem-solution; rarely peer reviewed |
| Compliance / audit report | Demonstrate adherence to standards | Regulators, accreditation bodies | Checklist-based; evidence-anchored; declarative prose |
| Incident / post-mortem | Analyse a failure or outage | Engineering leadership | Timeline; root cause; corrective action plan |
Canonical structure · 1 of 4
Poor
“An Interesting Study”
Adequate
“A Comparative Evaluation of Transformer-Based Models for Code Summarisation on the TLCodeSum Dataset”
Background (1–2 sentences) — the domain and the gap
Objective (1 sentence) — the specific aim
Methods (1–2 sentences) — how the aim was addressed
Results and conclusion (2–3 sentences) — principal findings and significance
Abstracts are indexed by digital libraries. No undefined acronyms, no citations to numbered references, no vague claims.
“This paper presents SecureVote, a blockchain-based voting protocol that achieves end-to-end verifiability while preserving voter anonymity. Our implementation on a permissioned Ethereum network processes 1,200 votes per second with a median confirmation latency of 1.4 seconds. Security analysis under standard cryptographic assumptions demonstrates resistance to double-voting, coercion and ballot tampering.”
Also in front matter: table of contents, list of figures, list of tables — generated automatically, three heading levels deep, dot leaders and right-aligned page numbers.
Canonical structure · 2 of 4
Diagnostic: if the reader cannot state your research question after reading it, the introduction has failed. Length: 1–1.5 pages in a conference paper; 15–25 pages across chapters in a thesis.
1 · Reproducibility. A competent researcher must be able to repeat the study from your description: equations, hyperparameters, hardware, software versions, dataset identifiers, evaluation protocol.
2 · Justification. Every choice carries a rationale citing prior art or a stated trade-off.
3 · Boundary conditions. State assumptions explicitly — “all experiments ran on a single NVIDIA A100 with CUDA 12.2”.
Sub-template: datasets · proposed model or algorithm · experimental setup (baselines, metrics, hyperparameters) · evaluation protocol (cross-validation, statistical tests, seeds) · implementation details
Canonical structure · 3 of 4
Pattern: topic sentence → table or figure → quantitative restatement. “Table 3 indicates that the proposed model achieves a 7.4 percent improvement in mean F1 across the ten folds (p < 0.01, Wilcoxon signed-rank).”
Commentary buried inside Results; and a Discussion that repeats the Introduction in different words. Test every discussion paragraph: does it add interpretive value not already in Results?
Conclusion is short (0.5–1 page): motivation, principal contribution, broader impact. No new data, no new citations. Back matter: references, appendices (proofs, full statistics, code), acknowledgements, author contributions and conflict-of-interest disclosures.
Canonical structure · 4 of 4
Subject–verb agreement across intervening phrases: “A collection of algorithms are presented” → “…is presented”
Parallel structure: “fast, reliable, and it is also easy to use” → “fast, reliable and easy to use”
Punctuation: serial comma; em dash for parenthetical emphasis; quotation marks for quotation only — italics for emphasis; semicolons between related independent clauses
Loose pronouns — “The compiler optimizes the loop, and it runs faster.”
Adverbial placement — “She said yesterday the system failed.”
Nominalisation — “The performance of an analysis of the data was undertaken” → “We analysed the data.”
A middle register: impersonal yet engaged, authoritative without condescension. “We” is accepted and often encouraged in CS. Terms such as throughput, latency, concurrency, idempotent, monotonic carry precise meanings — never their general-language sense.
Avoid gendered pronouns for generic referents — prefer singular they, “the user”, or restructuring. Avoid constructions that perpetuate stereotypes or exclude groups.
Referencing · 1 of 2
Citation in the text · bibliography entry in the list · style, the rules connecting them — confusing the three is why student reference lists fail internal consistency. Citation attributes credit, anchors claims to prior work, protects against plagiarism accusations, and lets readers verify and build on the work.
| Style | In-text marker | Reference list | Common in |
|---|---|---|---|
| IEEE | [n] | References, ordered by first citation | CS, EE — the de facto standard in computing |
| ACM | [n] | References, alphabetical (acmart class) | CS |
| APA 7 | (Author, year) | References, alphabetical | Psychology, education, CS education |
| Chicago (notes) | Footnote | Bibliography | Humanities |
| MLA 9 | (Author page) | Works Cited | Literature, some linguistics |
| Vancouver | (n) | References, by first citation | Medicine, biomedical journals |
| Harvard | (Author, year) | Reference list, alphabetical | Multidisciplinary, many local variants |
Referencing · 2 of 2
Original wording, quotation marks, page number. Rare in technical writing.
The idea restated in your own words, with citation. The norm in CS.
Condenses a longer work; cite the source as a whole.
Narrative — “Vaswani et al. (2017) demonstrate that…” emphasises the author.
Parenthetical — “…has been demonstrated (Vaswani et al., 2017)” emphasises the finding.
APA 7 shortens 3+ authors to first author + et al. IEEE and ACM list all authors up to a limit (commonly six).
Author(s) · year · title · container (journal, proceedings, book) · volume, issue, pages · publisher and location · identifier — DOI, arXiv ID, ISBN
Devlin, J., Chang, M., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. Proceedings of NAACL-HLT 2019, 4171–4186. https://doi.org/10.18653/v1/N19-1423
Zotero (open source, one-click browser capture) · Mendeley (freemium, PDF reader, institutional groups) · EndNote (commercial, team collaboration) · BibTeX / biblatex (native to LaTeX, highly portable)
Reproducible workflow: one .bib file, bibliography generated at compile time, a chosen .bst style.
Visual aids · 1 of 2
the reader must read precise values, the data are textually heterogeneous, or the comparison is between discrete categories.
the reader must apprehend a trend, a distribution, or a spatial relationship.
| Chart type | Best for | Cautions |
|---|---|---|
| Line | Time series, continuous trends | Avoid with more than about five series |
| Bar | Comparing quantities across categories | Sort by value unless the axis is time |
| Histogram | Distribution of a continuous variable | Specify bin width; not a bar chart |
| Box plot | Statistical dispersion across groups | Requires statistical literacy from the audience |
| Scatter | Relation between two continuous variables | Overplotting — use alpha or jitter |
| Heatmap | Matrix data: correlations, confusion matrices | Use a perceptually uniform colormap (viridis) |
| Sankey / network | Flows; relational data | Never Sankey for non-flow data; force-directed layouts mislead |
| Confusion matrix | Classification evaluation | Annotate with both raw counts and normalised values |
Visual aids · 2 of 2
booktabs convention.pdf, .svg)Use a structured environment (algorithm, algorithmic). Number every line for cross-reference. State input, output and a complexity comment.
Syntax-highlighted monospace (listings, minted). Excerpts only — the part that is the contribution. Specify language and version.
Sequence (inter-process), UML class (OO design), architecture or block (composition), ER (databases). Stick to standard notation — a non-standard diagram is worse than none.
Putting it together
1 · Title page — project title, student, supervisor, programme, date
2 · Abstract — background, objective, methods (Raspberry Pi benchmarks), result (15% energy reduction), significance
3 · Table of contents
4 · Introduction — growth of edge computing, the energy challenge, prior scheduling literature, the stated gap
5 · Methodology — hardware specs, baseline and proposed scheduler, energy measurement via INA219 sensor, workload suite
6 · Results — tables of energy, latency, throughput; figures of latency–energy Pareto fronts
7 · Discussion — trade-off interpretation, comparison to the Round-Robin baseline, limitations of single-board scope
8 · Conclusion and future work
9 · References — IEEE style
10 · Appendices — full measurement logs, code, datasheets
Title–content mismatch
Methodology gap — cannot be replicated
No clear research question
Results–discussion confusion
Inconsistent reference formatting
Purposeless visuals
Overgeneralising from n = 10
Structural plagiarism — too-close paraphrase
Errors in the abstract — read most closely
Consolidation
Take one paragraph from a recent paper in your own area, revise it with these principles, and submit original, revision, and a 200-word justification of your editorial choices.
Strunk & White (2000) · Minto (2009) · Schimel, Writing Science (2012) · Tufte (2001) · Whitesides, “Writing a Paper” (2011) · IEEE Editorial Style Manual · ACM Style Guide
Next — Lecture 2: From Research Idea to Complete Proposal: the problem, objectives, hypotheses, significance, scope and feasibility, and the ten components of the proposal document.