CSEG3060 · Research Methodology in Computer Science
Objectives, scope and significance in research
Ethical research begins in the library, not in the laboratory.
Unit III · Lecture 1 of 8 · 60 minutes · 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 · CO2
Define a literature study and separate it from general or casual reading
Articulate the principal objectives a literature study must achieve
Describe its scope across sources, time, disciplines and methods
Explain its significance as the foundation of rigorous, ethical research
Recognise how literature studies connect to research ethics — the Unit III theme
Foundational definition
Without method it becomes a casual bibliography: sources accumulate, nothing is decided.
Without appraisal it only paraphrases: every source is treated as equally authoritative.
Without synthesis it advances nothing: five summaries are still five summaries.
Where the evidence lives
IEEE Transactions, ACM Computing Surveys, JMLR.
NeurIPS, ICSE, SIGCOMM, CVPR. In computing these are archival, not preliminary.
Books, monographs, and doctoral theses — the most exhaustive reviews available on a narrow topic.
Microsoft Research, Google DeepMind. Results of real consequence outside conventional venues.
ISO, IEEE, NIST, W3C specifications — normative wording that must be quoted, not paraphrased.
arXiv, SSRN, white papers. Indispensable for currency; unrefereed.
Objectives · 1 of 3
Map the theories, algorithms, architectures, datasets, benchmarks and findings that already exist. Without this map you cannot legitimately claim to address an open problem.
The review is the mechanism by which gaps surface — and the articulation of a gap is itself a contribution.
| Gap type | What is missing | CS example |
|---|---|---|
| Knowledge | An unanswered empirical or conceptual question | No study measures energy cost of this layer |
| Methodological | No adequate method, tool or metric | No benchmark exists for the task |
| Theoretical | No framework explains a class of observations | No account of why scaling helps here |
| Population / context | An under-studied dataset, domain or region | Never evaluated on non-English corpora |
| Replication | Prior findings never independently validated | Headline ML result never reproduced |
Objectives · 2 of 3
Prevents costly repetition of published work — acute in computing, where architectures and benchmarks proliferate faster than anyone can informally track.
Reading reshapes the question: narrowing, broadening, re-framing, or operationalising it in measurable terms.
The literature is a repository of validated designs, metrics, statistical tests, simulators and tooling.
Before reading
“Can AI detect disease?”
After reading
“Does a 3D U-Net trained on MIMIC-CXR improve segmentation Dice over 2D baselines?”
Objectives · 3 of 3
Concepts, taxonomies and theoretical lenses make new observations interpretable rather than anecdotal.
Engaging with foundational work signals that the research was not conducted in intellectual isolation. Examiners expect this lineage.
Wide reading familiarises you with attribution norms. A researcher who has read widely is far less likely to plagiarise.
| # | Objective | Core function |
|---|---|---|
| 1 | Establish the state of the art | Map existing knowledge |
| 2 | Identify research gaps | Justify novelty |
| 3 | Avoid duplication | Ensure efficiency and originality |
| 4 | Refine problem and hypotheses | Improve precision |
| 5 | Select methods | Borrow and adapt validated tools |
| 6 | Build conceptual frameworks | Provide theoretical grounding |
| 7 | Establish scholarly credibility | Demonstrate engagement |
| 8 | Support ethical practice | Reinforce norms of attribution |
Scope · 1 of 2
Primary — original research articles, full conference papers, patents
Secondary — surveys, reviews, meta-analyses, textbooks
Tertiary — ACM DL, IEEE Xplore, Scopus, Web of Science, DBLP
Grey — arXiv, SSRN, technical reports, dissertations
Standards — ISO, IEEE, ACM, NIST, W3C
Foundational — Turing 1950, Dijkstra 1959
Contemporary — the last 3–5 years: current debates, datasets, benchmarks
Emerging — preprints signalling future direction
60–70% of citations from the last five years, while deliberately retaining the classics that anchor your lineage.
Common error: relying only on secondary sources. Citing a survey's characterisation of a paper rather than the paper itself propagates any error the survey made — and examiners can detect it.
Scope · 2 of 2
A review of deep learning for medical image segmentation must traverse:
Because interdisciplinary literatures are unbounded, state which neighbouring fields are in scope and which are excluded, with reasons.
Represent the field's methods — theoretical analysis, simulation, controlled experiment, case study, design science — so your own design is justified by contrast or extension, not by default.
Geographical bias — over-reliance on Anglo-American literature
Linguistic bias — excluding significant non-English work
These shape whose problems count as important and whose datasets count as representative.
Significance
The foundation for every later stage. Without it the design is poorly motivated, the contribution unclear, the evaluation criteria unjustified.
Builds critical reasoning, synthesis, disciplinary vocabulary, and the capacity to position your work in a live conversation.
Aggregated across a community, reviews identify trends, controversies and frontiers, and give newcomers a defensible entry point.
As computing governs decisions, surveillance, healthcare and transport, the review is how researchers meet the social, legal and ethical consequences of their work.
A review of facial recognition that never engages the literature on demographic error-rate disparity has met a technical standard while failing a societal one.
Unit III thesis
Ethical research practice begins long before data collection. This unit pairs literature studies with ethics because the two are causally linked — not merely adjacent.
Attribute ideas correctly and avoid plagiarism.
Recognise prior contributions; respect intellectual property.
Identify ethical controversies already debated in the literature.
Avoid the redundant — and ethically questionable — repetition of completed studies.
Mechanism, not exhortation: plagiarism is frequently a symptom of insufficient reading. A researcher who cannot restate a concept in their own words has not yet understood it.
Codified in the ACM Code of Ethics and Professional Conduct (2018) and the IEEE Code of Ethics · developed further in Lectures 4–8 of this unit.
Practical application
| Type | Purpose | Typical use |
|---|---|---|
| Narrative | Broad overview of a topic | Thesis background chapter |
| Systematic | Exhaustive, protocol-driven synthesis | Mapping a whole evidence base |
| Meta-analysis | Statistical synthesis of quantitative results | Pooling effect sizes |
| Scoping | Maps breadth and nature of activity | Emerging or heterogeneous fields |
| Integrative | Combines empirical and theoretical work | Mature fields needing synthesis |
| Umbrella | Synthesises multiple systematic reviews | Policy or curriculum decisions |
A B.Tech or PG project usually embeds a narrative or integrative review in a background chapter — but elements of systematic method (predefined search strings, explicit inclusion criteria) are increasingly expected.
Practical application
| Stage | Pitfall that bites here |
|---|---|
| Define scope | Bibliographic shopping — no integrating argument |
| Search | Closed-system searching — Google Scholar alone |
| Screen / assess | No critical appraisal — all sources equal |
| Extract / synthesise | Chronological listing instead of themes |
| Extract | Unmanaged references — no Zotero or Mendeley |
| Report | Recency bias · disciplinary parochialism |
| Report | Plagiarism through paraphrase — no attribution |
Screening commonly forces you back to the search; synthesis regularly reveals the need to search again. Returning to an earlier stage is evidence of a maturing review, not of earlier failure.
Install Zotero or Mendeley this week. Retrofitting citations at submission time is where most accidental attribution failures occur.
Case study
Recent surveys optimise accuracy under privacy constraints but neglect communication efficiency, energy consumption, and fairness across hospital sites with heterogeneous data.
A communication-efficient aggregation algorithm with formal differential-privacy guarantees, evaluated across heterogeneous clinical sites — a claim now warranted by the review.
The review must also surface literature on patient consent, secondary use of health data and bias in clinical algorithms — binding the project to the ethical theme of Unit III.
Your turn. For “Blockchain-based Trust Management in IoT Networks”, specify (a) three source categories, (b) two foundational works, (c) one likely research gap, (d) one ethical concern the review should surface.
Consolidation
Which single objective does your own draft review currently fail?
Webster & Watson (2002) · Creswell & Creswell (2018) · Fink (2019) · vom Brocke et al. (2009) · ACM Code of Ethics (2018)
Next — Lecture 2: Approaches to Literature Review: systematic, narrative, scoping and meta-analytical methods.