CSEG3060 · Research Methodology in Computer Science
Paraphrasing, citation styles, reference management and similarity checking
Knowing that plagiarism is wrong prevents nothing. A workflow does.
Unit III · Lecture 5 of 8 · 60 minutes · CO3
Dr. Mohsin Furkh Dar · School of Computer Science, UPES
→ / Space next · ← back · O overview · N speaker notes · F full screen
Learning objectives · CO3
LO1 · Understand — differentiate acceptable paraphrasing from disguised copying
LO2 · Apply — use APA 7th and IEEE citation styles correctly
LO3 · Apply — organise sources and generate bibliographies with Mendeley or Zotero
LO4 · Analyse — interpret Turnitin / URKUND similarity reports and act on them
LO5 · Create — formulate a personal anti-plagiarism workflow for your thesis
Orientation
Restating ideas in your own voice while preserving meaning.
The formal language that credits sources unambiguously.
Systematic organisation of bibliographic data.
The technological verification layer that confirms originality.
| Term | Meaning |
|---|---|
| Paraphrase | Restatement in different words and structure, meaning preserved |
| Citation | In-text reference acknowledging the source of an idea or finding |
| Similarity index | Percentage of submitted text matching comparison databases |
| Originality report | Annotated output showing matched sources and overlap statistics |
They reinforce each other: annotation during reading feeds paraphrasing; the reference manager supplies the citation; the similarity check verifies the result.
Target
| Type | Description | Severity |
|---|---|---|
| Direct | Verbatim copying without quotation marks or citation | Most severe |
| Mosaic / patchwork | Mixing phrases from multiple sources without citation | Severe |
| Idea | Using another's conceptual framework without attribution | Severe |
| Self | Reusing one's own published work without disclosure | Moderate to severe |
| Accidental | Unintentional misuse through poor note-taking | Moderate |
Accidental → reference management
Mosaic → genuine paraphrasing
Direct → quotation and citation
Idea → attributing concepts, not only sentences
CS papers contain algorithms, pseudocode and architectural diagrams. Copying an algorithm description and presenting it as your own contribution is plagiarism even if the words are modified — the idea belongs to its author.
Pillar 1 · part 1
Step 2 is the load-bearing one. Paraphrasing with the source in view produces patchwriting almost automatically, because the eye supplies the original syntax before the mind supplies an alternative.
Pillar 1 · part 2
“Deep learning is a particular kind of machine learning that achieves great power and flexibility by learning to represent the world as a nested hierarchy of concepts, with each concept defined in relation to simpler concepts.”
“Deep learning is a specific type of machine learning that achieves high power and flexibility by learning to represent the world as a nested hierarchy of concepts, where each concept is defined in relation to simpler ones.”
Critique: only synonyms changed; sentence structure intact. Turnitin will flag it.
“By modeling reality through layers of increasingly abstract representations — where each higher-level construct is built upon the lower-level ones — deep learning derives its representational capacity from hierarchical feature composition, distinguishing it from classical machine learning paradigms (Goodfellow et al., 2016).”
Critique: structure inverted, vocabulary domain-appropriate, meaning preserved, citation included.
Pillar 1 · part 3
| Technique | When to use | Length | Citation requirement |
|---|---|---|---|
| Quotation | The original wording is uniquely authoritative or stylistically essential | Exact words; usually ≤ 40 words | Required plus quotation marks |
| Paraphrase | The idea matters but the wording does not | Approximately the original's length | Required (no quotation marks) |
| Summary | A long passage must be condensed | Substantially shorter | Required (no quotation marks) |
Reserve direct quotation for definitions, formal theorems and standards language. Use paraphrase for algorithm descriptions and prior-work summaries.
Paraphrasing a shall clause from an IEEE standard or an RFC can change its normative meaning. Quote it.
Pillar 2 · part 1
Recent work has demonstrated the efficacy of transformer architectures for low-resource NLP tasks (Devlin et al., 2019).
“The model architecture is a multi-layer bidirectional Transformer encoder” (Devlin et al., 2019, p. 4171).
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. Proceedings of NAACL-HLT, 4171–4186. https://doi.org/10.18653/v1/N19-1423
Recent work has demonstrated the efficacy of transformer architectures for low-resource NLP tasks [1].
[1] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of deep bidirectional transformers for language understanding,” in Proc. NAACL-HLT, 2019, pp. 4171–4186, doi: 10.18653/v1/N19-1423.
Author set, title, venue, year, pages and DOI appear in both. Only the ordering, punctuation and in-text marker differ. Learning a style is learning a convention, not new content.
Pillar 2 · part 2
| Element | APA 7th | IEEE |
|---|---|---|
| In-text marker | (Author, Year) | [Number] |
| Reference ordering | Alphabetical by author | Order of appearance |
| Article title case | Sentence case | Lower case except proper nouns |
| DOI format | https://doi.org/… | doi: 10.… |
| Author name format | Last, F. M. | F. Last |
IEEE — IEEE, ACM, Springer LNCS and Elsevier submissions, and most B.Tech/M.Tech theses in Indian universities including UPES.
APA — HCI, cognitive science, education research, interdisciplinary work.
ACM Reference Format — for ACM venues specifically.
Beginning in APA and switching to IEEE mid-document produces a mixed bibliography that IEEE reviewers reject outright. Fix the style at the start, configure the reference manager to it exclusively, and verify against the official manual before submission.
Pillar 3 · part 1
| Feature | Mendeley | Zotero | EndNote |
|---|---|---|---|
| Developer | Elsevier | Roy Rosenzweig Center (open source) | Clarivate |
| Cost | Free, optional premium | Free, open source | Paid institutional licence |
| PDF storage | Excellent | Excellent (with web plugin) | Good |
| Word integration | Yes | Yes | Yes |
| Browser capture | Web Importer | Connector | Web capture |
| Collaboration | Private groups | Group libraries | Shared libraries |
| Best for | PDF-heavy workflows | Open-source advocates | Large institutional projects |
Mis-ordered numeric citations · inconsistent author formats · missing DOIs · orphaned in-text markers · forgotten references — the mechanical causes of accidental plagiarism.
It does not decide whether a citation is needed. That judgement remains yours.
Pillar 3 · part 2
Author_Year_Keyword_Title.pdfMendeley and Zotero export .bib files consumed by BibTeX (legacy, universal) or biblatex (modern, flexible). The Better BibTeX Zotero plugin generates stable keys such as vaswani2017attention.
The double braces around {BERT} stop BibTeX lowercasing the acronym under IEEE style — a reminder that the tool is a program, not a formatter.
Pillar 4 · part 1
Billions of continuously updated pages
Journal articles, conference proceedings, e-books
Prior student submissions at the same university
Submissions are broken into overlapping text segments; each is matched against the databases; a similarity index is computed as matched text over total length; colour-coded highlights link each match to its suspected source. Integrated into the UPES LMS.
URKUND was rebranded Ouriginal after a 2020 merger with PlagScan. MOSS remains the tool for source code; free web checkers offer basic reports with no reference exclusion.
Pillar 4 · part 2
UPES permits 10–20% for theses, excluding references, quotations and common methodological phrasing. A 0% score does not guarantee originality — translated, well-paraphrased and non-textual content (code, diagrams) can evade detection.
27% report. 8% cited quotations · 12% methodological phrasing common to all CS theses · 5% reference-list titles · 2% one uncited paragraph from a survey paper.
Effective index: 2%. The thesis is in good standing; the one paragraph must be rewritten with attribution.
Case studies
Original (McMahan et al., 2017): “Federated learning is a machine learning setting where many clients collaboratively train a model under the coordination of a central server, while keeping their training data decentralized.”
Student draft: “Federated learning is a setting of machine learning where many clients train a model collaboratively under the coordination of a central server, while keeping training data decentralized.” — structure preserved, trivial reordering: patchwriting.
Corrected: “The federated learning paradigm distributes model training across numerous client devices, each retaining local data, while a central coordinator aggregates updates without direct access to the underlying datasets (McMahan et al., 2017).”
A student begins coursework in APA, switches to IEEE for the thesis, and submits a mixed bibliography. IEEE reviewers reject it immediately. Resolution: fix the style at the outset, configure the reference manager, verify before submission.
Read and annotate → close the source → draft the paraphrase → insert the citation from the plugin → run a pre-submission check → compute the effective index → back up.
Verification is last, never first. A workflow that begins with a similarity check is editing for a score rather than writing with integrity.
Consolidation
Expected items: the three Turnitin databases · components of an IEEE in-text citation and reference entry · the UPES maximum similarity index excluding references and quotations · the five-step paraphrasing method with a CS example · Mendeley versus Zotero · the applied task (two paraphrased passages with IEEE citations and a justification) · the constructive task (design and justify a personal anti-plagiarism workflow).