M3 · Lesson 1 — Documenting Findings

The Paper
Summary Template

Reading without documenting is like cooking without a recipe.
You get it right once — and can never reproduce it.

01
M3 · L1 — The Problem

Why documentation matters

You will forget
what you read

"Six months later you remember reading it, but not what it said — and you can't find the insight you need for your own paper."

A systematic template solves this by forcing you to:

  • Extract the core idea in your own words
  • Record your critical observations while they're fresh
  • Note relevance to your own work immediately
  • Build a searchable knowledge base over time
02
M3 · L1 — The Template

One page per paper

The 6-field
summary template

Problem
What gap does this paper address?
Approach
What is their core method in one sentence?
Key Idea
The single insight that makes this work
Strengths
What does it do genuinely well?
Weaknesses
What are the real limitations?
Relevance
How does this connect to my work?
03
M3 · L1 — Worked Example

Applied to K-RagRec

What a filled
template looks like

Problem
LLM-based RS suffers from hallucination and lack of domain knowledge. Vanilla RAG introduces noise and ignores KG structure.
Approach
Retrieve hop-field KG subgraphs, filter by popularity, re-rank, then encode with GNN before injecting into LLM prompt.
Key Idea
Multi-hop KG indexing + GNN encoder allows structured graph knowledge to be efficiently injected into frozen LLMs.
Strengths
Efficient (near-direct inference speed). Strong hallucination reduction. Generalises zero-shot.
Weaknesses
No traditional RS baselines. Weak negative sampling evaluation. Requires KG availability.
Relevance
Relevant if extending LLM-RS with structured knowledge. GNN encoding approach potentially transferable.
04
M3 · L1 — Each Field

Why each field exists

Every row has
a purpose

Problem
Forces you to separate what they claim to solve from what they actually solve — often different.
Approach
If you can't write this in one sentence, you haven't understood the paper yet.
Key Idea
The single transferable insight — often the most useful thing when writing your own related work.
Strengths
What to cite positively. What to build on.
Weaknesses
Your research opportunities live here. A weakness you can fix is a paper waiting to be written.
Relevance
Forces an explicit connection to your work. Without this, notes become a graveyard of unconnected readings.
05
M3 · L1 — Quality Check

The most common mistake

Summarising vs
understanding

Bad summaries copy the abstract. Good summaries use your own words and your own judgement.

❌ Bad summary (copied)

"We propose K-RagRec, a novel framework that retrieves high-quality and up-to-date structure information from the knowledge graph to augment recommendations."

This is just the abstract. Zero understanding demonstrated.

✅ Good summary (understood)

"The core bet is that graph structure (not just text) is what LLMs need for RS. They use multi-hop GNN indexing to encode neighbourhood context — removing this drops accuracy by 45%, so the GNN is the actual contribution."

Shows the reader understood and evaluated the paper.

06
M3 · L1 — When to Document

Timing matters

Write during
Pass 2, not after

The best moment to document is while reading — not the day after. Observations decay fast.

  • Open your notes file before you open the paper
  • Fill each field as you reach that section
  • Write Weaknesses and Relevance last — after full Pass 2
  • Target: 15–20 min total per paper

RULE OF THUMB

If you can't fill the Relevance row, either the paper isn't relevant — or you haven't thought hard enough about why you're reading it.

You already have a template — the Excel file from M2 has all 6 fields ready. Paper 01 is waiting for K-RagRec.

07
M3 · L1 — Key Takeaways

What to remember

01

One page per paper

Six fields. Done during Pass 2. Written in your own words — not copied from the abstract.

02

Weaknesses = opportunities

The weaknesses row is where your next research idea lives. Never leave it empty.

03

Relevance is the point

A note without a relevance connection is an island. Force yourself to draw the line to your own work.

Next: M3 · L2 — The Literature Matrix

08
← → arrow keys to navigate
1 / 8