M6 · Lesson 3 — Writing Research Papers

Related Work
That Positions

Related work is not a history lesson.
It is an argument for why your paper needed to exist.

01
M6 · L3 — List vs Position

The fundamental difference

Listing vs positioning

❌ List-style (adds no value)

"MF was proposed by Koren et al. (2009). Neural CF was proposed by He et al. (2017). LightGCN simplified GCN for RS (He et al., 2020). Recently, LLM-based RS has emerged (Zhao et al., 2024)."

This is a bibliography. It tells the reviewer nothing about why your paper needed to exist.

✅ Positioning (earns its space)

"Graph-based methods (LightGCN, NGCF) capture collaborative structure but lack semantic knowledge. LLM-based methods (KAPING, G-Retriever) add semantics but treat items as text, ignoring their relational structure. Neither addresses the interaction between structure and semantics — which we resolve."

The rule: Every paragraph in your related work must end with "but none of these address X — which we solve." If a paragraph doesn't do this, it's padding.

02
M6 · L3 — Organisation

How to structure related work

Organise by theme,
not chronology

Group papers by the limitation they share, not the year they appeared. Typical RS related work sections:

  • Group 1: CF-based methods (MF, GCN, LightGCN) — limitation: no semantic knowledge
  • Group 2: KG-based methods (KGAT, KGCN) — limitation: no language understanding
  • Group 3: LLM-based methods (TALLREC, P5) — limitation: no structural knowledge
  • Group 4: KG-RAG methods (K-RagRec, KAPING) — limitation: your gap goes here
Paragraph template

One paragraph per group

Describe what the group does well → what they all share as a limitation → transition to the next group that partially addresses it → but still doesn't solve your specific problem.

Each paragraph is a stepping stone that narrows the gap until only your method fills it. The reader should feel the gap tightening with each paragraph.

03
M6 · L3 — Citing Fairly

Citation ethics and strategy

How to cite without
misrepresenting

  • Don't strawman: Describe prior work at its strongest, then show why even the strongest version doesn't solve your problem.
  • Cite original papers: If you're describing MF, cite Koren 2009 — not a survey that mentions MF. Reviewers notice when you cite surveys to avoid engaging with originals.
  • Cite concurrent work: If a preprint appeared during your submission, mention it. Not citing visible concurrent work looks like you're hiding it.
"Acknowledge strength. Identify limitation. Transition. Your paper should feel inevitable by the time the reader reaches your method."
04
M6 · L3 — Key Takeaways

What to remember

01

Position, don't list

Every paragraph ends with "but none address X — which we solve." No exceptions.

02

Theme over chronology

Group by shared limitation. Let the gap tighten across paragraphs until only you fill it.

03

Cite at full strength

Describe prior work at its best. A strawmanned related work section signals weak confidence in your own contribution.

Next: M6 · L4 — Writing the Method Section

05
← → arrow keys to navigate