Journal Q1 Polish
Purpose: Final polish pass before submitting to Q1 journals (ISI/Scopus indexed). Ensures paper meets top-tier standards for notation consistency, language quality, and experimental rigor.
When to use: After paper draft is complete, before submission. NOT for initial writing — use
for that.
Links to:
- — structure, sections, flow
- — diction, terminology, IEEE/ACM style
- — results tables, metrics format
- — self-review checklist
- — venue selection, submission prep
Step 1 — Notation & Complexity Sync (Paper ↔ Thesis)
Paper phải dùng notation và complexity format giống thesis để tránh mâu thuẫn khi defend.
1.1 Notation Table
Tạo bảng notation mapping giữa paper và thesis:
| Symbol | Paper | Thesis | Status |
|---|
| Learning rate | $\eta$ | $\eta$ | ✅ Match |
| Batch size | $B$ | $B$ | ✅ Match |
| Model params | $\theta$ | $\theta$ | ✅ Match |
Checklist:
1.2 Complexity Notation
Q1 journals expect O(·) notation with explicit assumptions:
Time Complexity: O(T · N · d)
T = number of rounds
N = number of clients
d = model dimension
Space Complexity: O(N · d)
Per-client model storage
Rules:
- Always state what each variable represents
- Include amortized complexity if relevant
- Compare with baseline complexity in the same format
1.3 Abstract ↔ Section 3.3 Complexity Sync
Critical for Q1: Complexity stated in abstract must exactly match the analysis in Section 3.3 (or equivalent methodology section). Mismatches are a common reviewer complaint.
Audit template:
Abstract claims: "O(T · N · d) time, O(N · d) space"
Section 3.3 states: "O(T · N · d) time, O(N · d) space"
Status: ✅ Match / ❌ Mismatch → fix: ___
Common mismatch patterns:
- Abstract says "linear time" but Section 3.3 shows O(N²)
- Abstract omits a factor (e.g., forgets communication rounds T)
- Abstract uses different variable names than Section 3.3
1.4 Sampling Ratio Notation Sync
When paper and thesis use different notation for the same concept, create explicit mapping:
| Concept | Paper | Thesis | Unified Form |
|---|
| Effective sampling ratio | | \min(s_{\text{config}}, T/N)
| \min(r_{\text{max}}, S/N)
|
Rule: Pick one form (prefer thesis notation if already established) and use consistently. Add a footnote in paper: "We use the notation from [thesis citation] for consistency."
1.5 Cross-Reference Audit
Step 2 — De-AI / De-translation Protocol
Q1 reviewers increasingly reject papers with AI-generated or machine-translated language. This step strips telltale signs.
2.1 AI Smell Detection (0-5 Score)
Audit your paper for these AI-generated patterns:
| Signal | Description | Example | Fix |
|---|
| Symmetrical structure | All bullets/paragraphs start with same word pattern | "Enhance X...", "Enhance Y...", "Enhance Z..." | Vary sentence openings |
| Abstract noun stacking | Chaining abstract nouns | "utilization of optimization strategies" | "using optimization" |
| Generic intro/outro | Vague opening/closing | "In today's rapidly evolving world..." | Start with specific problem |
| Excessive hedging | Too many qualifiers | "It could potentially be argued that..." | State directly: "X shows..." |
| Repetitive paraphrasing | Same idea restated differently | "This is important. This matters. This is significant." | State once, move on |
| Triple adjective stacking | 3+ adjectives before noun | "novel comprehensive robust framework" | Pick one: "robust framework" |
| Passive voice overuse | >50% passive sentences | "It was found that..." | "We found..." |
| Connector overuse | furthermore, moreover, additionally every paragraph | "Furthermore, ... Moreover, ..." | Vary or restructure |
Scoring:
0/5 — Natural human writing
1/5 — Minor AI痕迹, light edit needed
2/5 — Noticeable patterns, targeted fixes
3/5 — Multiple signals, significant revision
4/5 — Heavy AI smell, major rewrite
5/5 — Obviously AI-generated, total rewrite
Target: 0-1/5 for Q1 submission.
2.2 Hype Words Blacklist
BAN these words/phrases entirely:
| Banned | Replace with |
|---|
| delve into | examine, investigate, explore |
| leverage | use, apply, employ |
| robust | reliable, stable, consistent |
| cutting-edge | current, recent, state-of-the-art |
| novel | (just delete — let the work speak) |
| groundbreaking | (delete) |
| paradigm | approach, framework, method |
| landscape | field, domain, area |
| tapestry | (delete) |
| meticulous | careful, thorough |
| furthermore | (use "Additionally" or restructure sentence) |
| moreover | (same) |
| in conclusion | (delete — just end the section) |
| it is worth noting | (delete — if worth noting, just state it) |
| comprehensive | complete, full, extensive |
| innovative | (delete or specify what's new) |
| significant improvement | improvement of X% (be specific) |
| state-of-the-art | current best, recent methods (cite them) |
| demonstrates | shows, indicates, confirms |
| facilitates | enables, allows, supports |
| enhances | improves, increases |
| utilizing | using |
| aforementioned | (delete — restructure) |
| subsequently | then, next, after |
| preliminary | initial, early |
2.2 Passive → Active Voice
Q1 journals prefer active voice. Scan for passive patterns:
Passive (bad):
The model was trained on the dataset.
Experiments were conducted to evaluate performance.
It was found that the method outperforms baselines.
Active (good):
We trained the model on the dataset.
We evaluated performance through experiments.
The method outperforms baselines.
Exception: Methods section can use passive for standard procedures ("The dataset was split into 80/20 train/test").
2.4 Quantify Hype with Experimental Data
Rule: When encountering hype words, replace with specific numbers from your results.
| Hype phrase | Bad (vague) | Good (data-driven) |
|---|
| "extremely fast" | "The method is extremely fast" | "The method converges in 43 rounds vs. 67 for FedAvg (35.8% reduction)" |
| "significantly better" | "Our method performs significantly better" | "Our method achieves 94.3% accuracy, outperforming the best baseline by 2.2 percentage points (p < 0.001)" |
| "absolutely robust" | "The approach is absolutely robust to noise" | "Accuracy degrades by only 0.8% when noise increases from 0% to 30%" |
| "vastly superior" | "Our framework is vastly superior" | "Our framework reduces communication cost by 35% while maintaining comparable accuracy" |
| "extremely efficient" | "The algorithm is extremely efficient" | "The algorithm runs in O(N log N) time, 2.3× faster than the O(N²) baseline" |
Workflow:
- Search document for: extremely, vastly, absolutely, incredibly, remarkably, substantially, considerably
- For each hit, locate the corresponding result in your experiments
- Replace with: metric + value + comparison/baseline
2.5 De-translation Patterns
If paper was translated from Vietnamese:
| Vietnamese structure | English fix |
|---|
| "The method has the ability to..." | "The method can..." |
| "In order to..." | "To..." |
| "Due to the fact that..." | "Because..." |
| "At the present time" | "Currently" / "Now" |
| "In the event that" | "If" |
| "Despite the fact that" | "Although" / "Despite" |
| "On a daily basis" | "Daily" |
| "Has the potential to" | "Can" / "May" |
2.6 Sentence Length Audit
- Target: 15-25 words per sentence
- Hard max: 35 words (split if longer)
- Check: any paragraph with 3+ consecutive sentences > 25 words → rewrite
Step 3 — Q1 Results Table Standard
Q1 journals expect rigorous experimental reporting. This is the #1 rejection reason for ML/AI papers.
3.1 Mandatory Columns
| Column | Required | Notes |
|---|
| Method | ✅ | Full name + citation |
| Metric(s) | ✅ | Primary metric bolded |
| Mean | ✅ | Arithmetic mean |
| Std Dev | ✅ | MANDATORY for ablation studies |
| # Seeds | ✅ | Minimum 3, recommend 5 |
| p-value | ⚠️ | Required if claiming "significant improvement" |
3.2 Ablation Study Format
⚠️ CRITICAL: Every cell in an ablation table must include mean ± standard deviation. No exceptions.
WRONG (rejection risk):
| Method | Accuracy |
|-----------|----------|
| Baseline | 85.2 |
| + Module A| 87.1 |
| + Module B| 88.3 |
RIGHT (Q1 standard):
| Method | Accuracy | # Seeds | p-value |
|------------|----------------|---------|----------|
| Baseline | 85.2 ± 0.3 | 5 | — |
| + Module A | 87.1 ± 0.4 | 5 | 0.002* |
| + Module B | 88.3 ± 0.2 | 5 | <0.001* |
* Statistically significant (paired t-test, α=0.05)
Rule: If a result cell shows only a single number (e.g.,
), it is incomplete. Every value must be
format.
3.3 Seed Count Justification
Include in experimental setup:
We report mean ± standard deviation over N independent runs with
different random seeds. We use [5/10/30] seeds to ensure reliable
statistical inference and reproducibility of our results.
Seed count guidelines:
| Domain | Minimum Seeds | Justification |
|---|
| Deep Learning (deterministic) | 3-5 | Low variance, GPU determinism |
| Deep Learning (stochastic) | 5-10 | Moderate variance |
| Federated Learning | 5-10 | Client sampling variance |
| Metaheuristic / Evolutionary | 30 | Central Limit Theorem (n ≥ 30 for normality) |
| Reinforcement Learning | 10-30 | High variance, environment stochasticity |
Why CLT matters for n ≥ 30:
- Sampling distribution of mean approximates normality regardless of population distribution
- Enables parametric tests (t-test, ANOVA) even for non-normal results
- Standard expectation in empirical research methodology
Template:
We report mean ± standard deviation over [N] independent runs
(seed ∈ {42, 123, 456, ...}). [N] seeds ensure [reliable statistical
inference / CLT normality / reproducibility] per standard empirical
methodology.
3.4 Statistical Tests
| Claim | Required Test |
|---|
| "outperforms" / "better than" | Paired t-test or Wilcoxon signed-rank |
| "comparable" / "similar" | Equivalence test or confidence interval overlap |
| "robust to hyperparameter" | Sensitivity analysis table |
3.5 Results Table Checklist
Final Polish Checklist
From
From
From
From
Integration Flow
paper-writing (draft complete)
↓
journal-q1-polish (this skill)
├── Step 1: notation-sync
├── Step 2: de-ai-protocol
└── Step 3: q1-results-standard
↓
internal-critique (final review)
↓
publication-strategy (venue selection + submission)