M3 · Lesson 2 — Documenting Findings

The Literature
Matrix

Individual paper notes tell you what one paper says.
The matrix shows you what the field hasn't done yet.

01
M3 · L2 — What Is It

From notes to structure

One row per paper,
one column per dimension

The literature matrix is a structured table where each row is one paper and each column is a research dimension:

  • Problem type
  • Method category
  • Datasets used
  • Evaluation metrics
  • Key limitation
"When you see 5 papers in a row with the same gap in the same column — that's your research opportunity."
02
M3 · L2 — Example

KG + LLM recommendation papers

A mini matrix

Paper Problem Method Dataset Cold-start? Trad. Baselines? Key Limitation
K-RagRec (2025) Hallucination + knowledge gap KG subgraph RAG + GNN ML-1M, ML-20M, Book ✓ Yes ✗ No Requires KG availability
G-Retriever (2024) LLM hallucination on graphs Graph RAG + LLM Various graphs ✗ No ✗ No Not RS-specific
KAPING (2023) LLM knowledge gap KG triple retrieval ML-1M ✗ No ✗ No Text only, no structure
LightGCN (2020) Over-parameterised GCN Simplified GCN Gowalla, Yelp ✗ No ✓ Yes No side info, no KG

Pattern visible immediately: The "Trad. Baselines?" column is almost entirely ✗. Three papers in the same sub-area all avoid comparing to classical CF. That's a systematic gap.

03
M3 · L2 — Gap Finding

The core skill

Gaps become
research questions

Once you have 5–10 papers in a matrix, scan each column vertically. Look for:

  • Consistent blanks — no one has addressed this
  • Consistent same answer — everyone uses the same dataset; untested on others
  • Gradual improvement — the field is converging; are you too late?
  • Contradictions — two papers claim opposite things about the same method
Paper
Cold-start
No-KG
Dense eval
K-RagRec
✓
✗ GAP
✗ GAP
G-Retriever
✗ GAP
✗ GAP
✗ GAP
KAPING
✗ GAP
✗ GAP
✗ GAP
LightGCN
✗ GAP
✓
✓

Yellow = gap. A column full of yellow = a research paper.

04
M3 · L2 — Column Design

Designing your matrix

Choose columns
based on your research area

Always include

Fixed dimensions

Problem type, Method category, Datasets, Metrics, Key limitation. These apply to every RS paper.

Add for your focus

Domain dimensions

Cold-start? Sequential? LLM-based? Graph-based? Knowledge-aware? Pick 3–4 that define your research space.

Add for positioning

Comparison dimensions

Does it compare to trad. CF? Does it test on sparse data? Does it evaluate fairness? These reveal community norms.

Your Excel file already has the Literature Matrix sheet with standard columns. Add 2–3 columns specific to your research focus — that's where your unique insight comes from.

05
M3 · L2 — Workflow

Making it stick

The matrix only works
if you maintain it

Simple workflow:

  • Fill one row immediately after each paper summary — while it's fresh
  • Review the full matrix once a month — look for emerging patterns
  • Before writing any paper section, sort by the relevant column
  • Add a row even for papers you reject after Pass 1 — record why
"The matrix pays back 10× when you're writing your related work section and need to position your paper in the field."

Instead of re-reading 20 papers, you scan the matrix in 5 minutes and immediately see where your work sits.

06
M3 · L2 — Key Takeaways

What to remember

01

Rows = papers,
Columns = dimensions

Fill one row per paper immediately after your summary. Consistency is everything.

02

Scan columns for gaps

A column with consistent blanks = a problem the field hasn't solved = your research opportunity.

03

Design columns around your focus

Standard columns track the field. Your custom columns track what matters to your specific research direction.

Next: M3 · L3 — Connecting Ideas Across Papers

07
← → arrow keys to navigate
1 / 7