how to write a good paper

Writing Perfect Papers_哔哩哔哩_bilibili

致敬李葆春老师

Part1. write a perfect paper

1. story => high level idea （tasteful design）=》 advance the state-of-the-art

• the problem
• what's new? your contribution & solution compared with the related work, solving the same problem.
  • if your problem is slightly different from the previous, prove/justify your problem is important, practical, and necessary.
• how is it better? how did you advance the state-of-the art
  • offer|prove better results than before by theoretical properties and better experiments results

---

---

2. work

• read and write in an interactive way
• reading => become an expert on the problem
  • how?-> from the single paper (high cited and seminal) , than do an expanding search -> finally, narrow down to the tiny problem
  • read quickly first, read more carefully if necessary later
• writing =>
  • write about what you understand, write every day
  • idea => write a paper (related work, what's challenges or what is hard or why the problem is important, without copying and pasting other papers, what needs to fix, what's new in the context of relate work) => do research
  • write in a crystal clear way => write about one problem and one solution.
  • keep your reads engaged

2.2. the structure of the paper

• title
  • reflects the essence of the new idea
  • as simple as short as possible
    • does not have to be a precise summary of the paper
    • does not need to include all the keywords
  • get the general idea of the paper at a glance
• abstract
  • convey the essential info about the paper: the three core components
  • structure
    • 1 sentence to state the background
    • 1 sentence to state the problem
    • 2-4 sentences to state the original, i.e., high-level and basic ideas of your work and contribution, compared with the state-of-the-art
    • 1 scents to state how the solution is better, using analyses, simulation or expriments.
• introduction

1. self-contained
2. don't spend too much space on the background and related work
3. don't make your work sound more original than it really is
4. structure

---

• ---

  General overview:  first para. (2-3 sen.) give the general overview of the research field => offer the basic facts to warm up the readers and prepare for the problem statement
• State the problem (challenge) and existing solutions from a high-level (1-2 para.)
  • include an intuitive/basic idea (example solution) to the problem and its existing solution intuitively (what does idea mean => the chief point the writer makes about a topic, which sums up the primary messages)
  • do not mix the basic idea with unnecessary details.
• Limitation of the existing solutions that motivates this paper (2-3 para.)
• Proposed solution: main idea (1-2 para)
•  Original highlights of the proposed solution (1-2 para) =>  What's new
  • why the proposed is different/better from/than the existing solution.
  • state 1-2 most impressive highlights
  • use bullets
• Attention

1. use a table to compare the important properties with the existing solution
2. use an simple example to show your main idea works in a special case
3. use a figure to illustrate your example, figures as much as you can
4. the flow of ideas

• each para. contains one complete idea
• don't write a very long abstract (200-250 works ), introduction => one page,
• related work (half a page), conclusion (1-2 para), background =>don't use a lot of space for unnecessary background, keep the motivation short and concis

  ---

---

 3. the art （details）

• use git，source code，references
• use spell checker by unix command-line tool (spell)
• proofreading
• use transitional words for sentences, paras, sections
• the comma in the middle of the sentence is one at most for each sentence,  to prevent it too long
• don't abuse where, in which, whose, so that, and such that?
• don't use words that too inform and colloquial(口语化）
  • colloquial, e.g., a lot of, big (use large, substantial instead)
• use oxford dictionary of english
• plural nouns (复数名词） are typically without an article (a/an/the)
• no need of the for section title
• the tools for drawing => draw.io (https://diagrams.net)
• find out the venue where the paper has been published, instead of a link to arXiv

---

---

---

Part2. write a good paper fast

1. know your ddl

2. don't follow, lead. allocate your time well => use gantt chart to manage your project

3. choose a problem with right size

•  make solid improvements over one paper, and one paper only
• the paper with source code and you enjoy reading, give up the paper if it needs long time to reproduce
• read paper with its source code to understand completely and deeply, then you will have to have a deadline for creating something new
• If the feasibility of your solution is promising, think about its theoretical properties - anything provably correct?
  • 是否能从理论层面解释为什么会有提升