论文写作课程总结

论文写作课程，顾名思义，为了帮助我们更加顺利的写论文，刚开始还觉得是不是开课太早了，现在离写论文还好远，学完这门课后感觉还是收获颇多，给了我们写论文一个很好的思路和例子。

下面是我的一些总结和体会~~

首先，要明白为什么要做研究、写论文？
对于学术界的大佬们来说，就是推动科学的进步、带动人类文明的发展！但是对于我们研究生来说，有些是为了达到毕业条件，针对自己研究方向的领悟都不深，但是在做研究和写论文的过程中，能够对于自己的研究有一定的进步以及对自己的研究工作更加深刻的认识。

其次，我们要针对什么方向发论文呢？
导师推荐的方向肯定是最好的选择，导师们在行业内摸爬滚打，经过了很多年的研究，对于各个方向都会有自己独特的见解，特别是对于导师自己的领域固然是很了解的，知道这个方向好不好，容不容易出成果，得到的成果是否有用的。比如有些方向可能上网搜了一下，没有人做这方面的研究，那很可能是没有研究意义，不能被我们轻易发现的研究方向怎么会没有人研究呢！！！这个时候一定要反思一下是不是自己的问题（bushi），如果这个方向确实是需要成果的但是目前还没有人在研究，那还愣着干啥，赶紧冲啊！

一、知识点总结

1、论文可以在别人总结的模板基础上写，这样写出来的论文比较符合要发的刊物的要求
2、一些从来没有在任何学术论文中出现的单词是不可以使用的，比如一些简写、一些形容词的应用，以及对于自己论文工作描述的时候，一些动词是不可以乱用的
3、如果两个同义词，使用https://www.linggle.com 查看单词或词汇的使用频率，用频率高者
4、注意大小写、空格、引号、数学符号等的应用
5、一个非常棒的英文句式网站 www.phrasebank.manchester.ac.uk
6、找一篇已经发表论文的 .tex 文件来学习数学表达式
7、如果论文涉及不少数学符号, 应该给出一个符号表, 便于读者查阅
8、不要对式子、符号进行额外的、特殊的处理，包括强行增加空格、花括号等
9、一般3-5个关键词，按字母表顺序排序
10、实验成果是要放在第一位的，实验成果有了，就有了90%的东西，后续写论文也就快了
11、 注重图注，很多审稿人不太看文章内容，主要看你的图是什么，做了什么
12、 兴趣是很重要的，有兴趣才有后续研究下去的动力
13、很多能想到的idea其实已经是旧大陆了，别人已经做过了，文献能找到，或者找不到的话，就是这个方向没有太大的意义
14、引言的话，就是要写清楚所做的工作，让审稿人读完之后，拥有对论文80%以上的判断，并且采用与摘要相同的节奏，每段5-10句，每段50-150个单词
15、在引言部分，如果采用计算机领域中流行的开局一张图，那么后续就应该围绕该图进行解释
16、引言的最后一段以 The rest of the paper is organized as follows，开头即可
17、在使用数学表达式的时候，理论应该完备，使用的符号要保持同一风格
18、结论部分不要太长，5句就够了，避免使用与摘要内容相同的句子，主要讲获得哪些观察和结论，如果要讲述进一步的工作，可以列出3-5条来
19、尽可能使用矢量图. 这样在放大的时候就不会失真，图注可以相当长，引用图片时, 应保持与图片编号的一致，如用 Fig. 1 对应 Fig. 1，但作为主语时, 建议使用 Figure 1
20、有些期刊要求提供 Graphical abstract, 即使用一张图 (含图注) 把论文的主要思想介绍清楚

二、关于Latex

通过论文写作课程才知道了Latex这个软件，感觉对于论文、特别是排版方面大有益处。
1、找出想要发表的期刊已经录用论文的源文件或者提供的模板
2、cls文件用于控制论文的总体格式，使用不同的格式文件时, .tex 源文件头部需要进行相应调整. 正文不一定调整, 除非涉及图、表、公式的排版
3、bst 文件用于控制参考文献的格式，如IEEE出版社就是IEEEtran.bst，Elsevier出版社就是elsart-num-sort.bst以及model5-names.bst，Springer出版社就是splncs.bst
4、使用 usepackage导入包
5、尽量不要引入特殊的包, 以免在其它系统 (特别是期刊投稿网站上) 上运行不出来
6、使用 \newtheorem 增加一些自动编号项
7、Latex提供了bib文件进行参考文献的管理，不要直接使用网上的bibitem

IEEEtran.cls 例：

\documentclass[journal]{IEEEtran}

\usepackage{algorithm}
\usepackage{algorithmic}
\usepackage{arydshln}
\usepackage{amsmath}
\usepackage{amssymb}
\usepackage{booktabs}
\usepackage{bm}
\usepackage{hyperref}
\usepackage{multirow}
\usepackage[misc]{ifsym}
\usepackage{pdfpages}
\usepackage{subfigure}
\usepackage{url}

\renewcommand{\algorithmicrequire}{\textbf{Input:}}
\renewcommand{\algorithmicensure}{\textbf{Output:}}
\newtheorem{definition}{Definition}

\hyphenation{algorithm}
\hyphenation{classifi-cation cha-llenge costs}
\hyphenation{values}

\begin{document}

\title{This is the title}

\author{Mei~Yang,
        Yu-Xuan~Zhang,
        Xizhao~Wang,~\IEEEmembership{Fellow,~IEEE},
        and~Fan~Min,~\IEEEmembership{Member,~IEEE}
\thanks{This work was supported by xxx.).
(\emph{Corresponding author: Fan Min}.)}
\thanks{Mei Yang is with the School of Computer Science, Southwest Petroleum University, Chengdu 610500, China
(e-mail: yangmei@swpu.edu.cn).}
\thanks{Yu-Xuan Zhang is with the School of Computer Science, Southwest Petroleum University, Chengdu 610500, China
(e-mail: 201921000434@stu.swpu.edu.cn).}
\thanks{Xizhao Wang is with Institute of Big Data, Shenzhen University, Shenzhen 518060, China
(e-mail:  xizhaowang@ieee.org).}
\thanks{Fan Min is with the School of Computer Science; Institute for Artificial Intelligence, Southwest Petroleum University, Chengdu 610500, China
(e-mail: minfan@swpu.edu.cn).}
}

\maketitle

\begin{abstract}
This paper xxx.
\end{abstract}

\begin{IEEEkeywords}
Distinguishability, ensemble learning, mapping, multi-instance learning and self-reinforcement.
\end{IEEEkeywords}

\IEEEpeerreviewmaketitle

elsarticle.cls 例：

\documentclass[preprint]{elsarticle}

\usepackage{rotating}
\usepackage{natbib}
\usepackage{multirow}
\usepackage{amsmath}
\usepackage{amssymb}
\usepackage{latexsym}
\usepackage{threeparttable}
\usepackage{algorithm}
\usepackage{algorithmic}
\usepackage{array}
\usepackage{graphicx}
\usepackage{epstopdf}
\usepackage{subfigure}
\usepackage{url}
\usepackage{bm}
\usepackage{color}

\newtheorem{property}{Property}
\newtheorem{example}{Example}
\newtheorem{proposition}{Proposition}
\newtheorem{lem}{Lemma}
\newtheorem{rem}{Remark}
\newtheorem{corollary}{Corollary}
\newtheorem{problem}{Problem}
\newproof{proof}{Proof}
\newtheorem{theorem}{Theorem}
\newdefinition{definition}{Definition}
\newtheorem{assumption}{Assumption}
\newtheorem{conj}{Conjection}
\DeclareMathOperator*{\argmax}{argmax}
\DeclareMathOperator*{\argmin}{argmin}
\DeclareMathOperator*{\sign}{sign}
\hyphenation{con-cept}
\hyphenation{ob-jects}

%\journal{Information Sciences}

\begin{document}
%\linenumbers
\begin{frontmatter}

\title{This is my title}

\author[1]{Author 1}
\author[2]{Author 2}
\author[1,2]{Fan Min\corref{cor1}}

\address[1]{School of Sciences, Southwest Petroleum University, Chengdu 610500, China}
\address[2]{School of Computer Science, Southwest Petroleum University, Chengdu 610500, China}

\cortext[cor1]{Corresponding author. Tel.: +86 135 4068 5200.\\ Email address: minfan@swpu.edu.cn (F. Min).}

\begin{abstract}
This paper introduced xxx
\end{abstract}
\begin{keyword}
Active learning, cost-sensitive learning, multi-label learning, missing labels.
\end{keyword}
\end{frontmatter}

svjour3.cls 例：

\documentclass[smallextended]{svjour3}
%\documentclass[twocolumn]{svjour3}
\hyphenation{al-go-ri-thms}
\hyphenation{re-sults}

%\usepackage[pdftex]{graphicx}
\usepackage{hyperref}
\usepackage{amsmath}
\usepackage{amssymb}
\usepackage{marvosym}
\usepackage{enumerate}
%\usepackage{amsthm}
\usepackage{graphicx}
\usepackage{algorithm}
\usepackage{algorithmic}
\usepackage{subfigure}
\usepackage{threeparttable}
\usepackage{multirow}
\usepackage{colortbl}
\usepackage{latexsym}
\usepackage{color}

\DeclareMathOperator*{\argmax}{argmax}
\DeclareMathOperator*{\argmin}{argmin}

\begin{document}
  \pagestyle{headings}

  \title{{\color{blue}Three-way} active learning through clustering selection}

%  \titlerunning{Lecture Notes in Computer Science}
  \author{Fan Min (\Letter) \and Shi-Ming Zhang \and Davide Ciucci \and Min Wang}
%  \author{Double-blind review}

\institute{F. Min\\
School of Computer Science, Southwest Petroleum University, Chengdu 610500, China\\
  \email{minfan@swpu.edu.cn}\\
  \\
S.-M. Zhang\\
School of Computer Science, Southwest Petroleum University, Chengdu 610500, China\\
  \email{zhangshiming@stu.swpu.edu.cn}\\
  \\
D. Ciucci\\
DISCo, University of Milano-Bicocca, viale Sarca 336/14, Milano 20126, Italy\\
  \email{davide.ciucci@unimib.it}\\
\\
M. Wang\\
School of Electrical Engineering and Information, Southwest Petroleum University, Chengdu 610500, China\\
\email{wangmin@swpu.edu.cn}
}
  \maketitle

  \begin{abstract}
This paper xxx.
\keywords{Active learning, clustering, granular computing, three-way.}
   \end{abstract

llncs.cls 例：

\documentclass[runningheads]{llncs}

\usepackage{graphicx}
\usepackage{subfigure}
\usepackage{times}
\usepackage{epsfig}
\usepackage{amsmath}
\usepackage{amssymb}
\usepackage{multirow}
\usepackage{color}
\usepackage[table]{xcolor}
\usepackage{overpic}
\usepackage{wrapfig,lipsum,booktabs}
\usepackage{url}
\usepackage{algorithm,algpseudocode}
\usepackage{algorithmicx}

\usepackage{epstopdf}
\usepackage{comment}
\usepackage{enumerate}

\graphicspath{{./figures/}}

\begin{document}
\title{This is my title}
\author{Chao-Fan Pan\orcidID{0000-0001-5345-2746} \and
  Fan Min\orcidID{0000-0002-3290-1036}
  \and Heng-Ru Zhang\orcidID{0000-0001-9187-9847}
}

\authorrunning{Chao-Fan Pan et al.}
\institute{School of Computer Science, Southwest Petroleum University, Chengdu 610500, China
  \email{pan.chaofan@foxmail.com, \{minfan, zhanghr\}@swpu.edu.cn}}

\maketitle

\begin{abstract}
  This paper proposes xxx..
  \keywords{Behavior imitation \and Imitation learning  \and Meta-learning.}
\end{abstract}

三、论文题目

1、论文的题目必须要有吸引力，不然审稿人看了没有兴趣的话，就哦豁了
2、题目要让别人看了能够理解在研究什么东西，且容易被检索到，论文被别人引用的次数也是很重要的
3、题目的字数要控制在40-60个字母之间，题目越长限定的东西也就越多，如果主要贡献力为某算法的话，题目的缩写就应该为该算法的名称
4、尽量不用based on，不然就说明你是基于别人方法的一个研究、一个简单的拓展或应用，创新性不够。使用 through, with 等来表示技术
5、多多分析顶刊论文的题目

四、论文摘要

1、说明研究的问题及其重要性
2、讲述已有的工作，以及其局限性，但局限性不要指责的过于强烈，要尊重别人的劳动成果
3、讲述本文的工作、实验结果以及提升等

下面引用一下老师给的例子：
例 1. Three-way active learning through clustering selection.
1、In clustering-based active learning, the performance of the learner relies heavily on the quality of clustering results.
第 1 句: 问题的重要性
2、Empirical studies have shown that different clustering techniques are applicable to different data.
第 2 句: 已有工作
3、We propose the three-way active learning through clustering selection (TACS) algorithm to dynamically select the appropriate techniques in the learning process.
第 4 句: 本文内容 (注意并没有第 3 句)
4、The algorithm follows the coarse-to-fine scheme of granular computing coupled with three-way instance processing.
第 4’ 句: 主要内容的补充说明
5、For label query, we select both representative instances with density peaks, and edge instances with the maximal total distance.
第 5 句: 第 1 项技术
6、For block partition, we revise six popular clustering techniques to accommodate the binary splitting.
第 6 句: 第 2 项技术
7、For clustering evaluation, we de ne weighted entropy
with 1-nearest-neighbor.
第 7 句: 第 3 项技术
8、For insufficient labels, we design tree pruning techniques with the help of a block queue.
第 7’ 句: 第 4 项技术
9、Experiments were undertaken on twelve UCI data sets.
第 8 句: 实验设置
10、The results show that TACS is superior to single clustering technique based algorithms and other state-of-the-art active learning algorithms.
第 9 句: 实验结果

例 2. Test-cost-sensitive attribute reduction
1、In many data mining and machine learning applications, there are two objectives in the task of classification: one is decreasing the test cost, the other is improving the classification accuracy.
理文编辑把两个句子合并, 导致比较长
2、Most existing research work focuses on the latter, with attribute reduction serving as an optional pre-processing stage to remove redundant attributes.
没写 However, 但挖了坑
3、In this paper, we point out that when tests must be undertaken in parallel, attribute reduction is mandatory in dealing with the former objective.
填坑
4、With this in mind, we posit the minimal test cost reduct problem which constitutes a new, but more general, difficulty than the classical reduct problem.
本质上相当于 First
5、We also define three metrics to evaluate the performance of reduction algorithms from a statistical viewpoint.
本质上相当于 Second
6、A framework for a heuristic algorithm is proposed to deal with the new problem; specifically, an information gain-based λ-weighted reduction algorithm is designed, where weights are decided by test costs and a non-positive exponent λ, which is the only parameter set by the user.
本质上相当于 Third
7、The algorithm is tested with three representative test cost distributions on four UCI (University of California-Irvine) data sets.
实验设置
8、Experimental results show that there is a trade-off while setting λ, and a competition approach can improve the quality of the result significantly.
结果
9、This study suggests potential application areas and new research trends concerning attribute reduction.
吹牛. 不过本文的引用次数确实达到了 300+

例 3. Three-way recommender systems based on random forest
1、Recommender systems guide their users in decisions related to personal opinions about items.
2、Most existing systems implicitly assume the recommender behavior as a binary classification.
3、That is, the incoming item is either recommended or not.
虽然没有用 However, 但有那个意思.
4、In this paper, we propose a framework integrating three-way decision and random forest to build recommender systems.
5、First, we consider both misclassification costs and teacher cost.
6、Misclassification costs are paid for wrong recommendation behaviors, while teacher cost is paid to consult the user actively for her tendency.
7、Second, with these costs, a three-way decision model is built and rational settings of α and β computed.**
8、Third, we build a random forest to compute the probability P that a user likes an item.
9、Fourth, α , β and P are employed for determining the recommender behavior to users.**
总共 4 个方面
10、The performance of the recommender is evaluated by the average cost.
11、Experiments results on the well-known MovieLens dataset show that the threshold pair ( α ∗ , β ∗ ) determined by three-way decision is optimal not only on the training set, but also on the testing set.

五、文献综述

1、每篇文章都有文献综述，表示对前人工作的尊重
2、可以在不同的地方描述
3、需要进行分门别类的介绍
4、不能将参考文献的引用作为句子的主语、宾语等
5、以年份为主线的综述也不值得提倡，也不能一次性引用过多的文献以及对文献进行全句引用

六、实验部分

1、数据集越多, 覆盖领域越广, 结果就越可信，一般12-20个公开数据集就够了，如果数据集很珍贵，可以使用人造数据集，或者数据集随机采样成多个
2、数据集大些更好. 如果是结构化数据, 有 10000个样本和 100 个属性就比较好，否则, 审稿人就会说你使用了玩具数据集
3、获取自己主要参考文献中的数据，有利于进行比较，你甚至不需要把他们的方案实现，在一些领域如图像和视觉, 会有一些专门的数据集供大家测试.
4、要么获得对比算法的源码，要么实现它们，由于使用平台不同，数据采样不同, 有时候你重现别人的实验，会发现结果不大一致，如果没有特别的原因，就使用自己实现的版本
5、结果的展示形式取图或表中的一种就行了，都展示会显得很重复
6、在实验的地方采用自问自答的方式，在实验之前提出这些问题，在实验结果列出之后逐个回答

例：在实验这一节开始的时候
In this section, we report the results of experiments to analyze the effectiveness of the TACS algorithm. Through the experiments, we aim to answer:
1、Is the TACS algorithm more accurate than popular supervised classification algorithms?
2、Is the TACS algorithm more accurate than popular active learning algorithms?
3、Is the TACS algorithm more accurate than single clustering technique based algorithms?
4、Can the TACS algorithm select appropriate base clustering techniques?

列完所有图表之后：
Now we can answer the questions proposed at the beginning of this section.
1、TACS is more accurate than popular supervised classification algorithms, including C4.5, NB, RF, etc. This is validated by Table 3. Unfortunately, on some datasets such as Ionosphere, it is significantly worse than some other algorithms such as RF. The reason may be that clustering techniques do not perform well on those datasets.
2、TACS is more accurate than popular active learning algorithms, including QBC, MAED, and KQBC. This is validated by Table 4. It was also defeated by MAED on the Heart dataset. The reason may be that for some datasets, informative instances are more important than representative ones.
3、TACS is more accurate than single clustering technique based algorithms. This is validated by Table 5. It is the best, or the second best one on all datasets.
4、In most cases, TACS can find out the appropriate base clustering techniques. This is validated by Table 6.

七、审稿意见

1、稿件投到期刊，如果编辑让你修改，多半就有戏了
2、思想上务必要端正态度，编辑和审稿人花大量时间义务审稿不是想为难作者，而是来帮助改进论文质量
3、问题应该直接回答, 而不要顾左右而言他
4、回复不要太长，正文中进行相应的修改才是重点，审稿人更关注论文正文修改得如何，毕竟它才是呈现给读者的内容
5、正文中修改部分应用蓝色字标出
6、审稿人会对论文进行一个总结, 可以直接如下所述回复表示感谢即可：
（1）Answer: Why is determining the number of labels in cost-sensitive active learning an important issue?
Response: In cost-sensitive active learning, a higher number of queried labels leads to higher teacher costs, but may also reduce the misclassification cost. Hence, the number of labels determines the trade-off between the teacher and misclassification costs. We have added some text to explain this in the second paragraph of the introduction.

（2）Answer: Why is the optimization objective of cost-sensitive active learning to minimize the sum of the teacher cost and the misclassification cost?
Response: Cost-sensitive learning usually aims to minimize the total cost. Because we consider teacher and misclassification costs in this paper, the optimization objective is to minimize their sum. We have added some explanations to clarify this in Section 3.1.

（3）Answer: Page 7, “Min et al. [23] considered the classification expectation and divided the entire data into three regions.” Do you use a similar method of three regions in this paper?
Response: In paper [23] (now [28]), Region I contains instances for which the expected misclassification cost is lower than the teacher cost, Region II contains instances to be labeled by the oracle, and Region III contains the remaining instances. In this paper, we use a similar method that consists of three regions. Region I contains instances labeled by the oracle, Region II contains instances classified by the active learner, and Region III contains the remaining instances.

（4）Answer: In Figure 1, “Round 1” may be better than “Step 1.”
Response: We accept the reviewer’s suggestion.

（5）Answer: The English writing should be polished, especially in the experimental part.
Response: We thank the reviewer for this important suggestion. We have consulted Edanz Editing to address the language issues.

（6）Answer: Experiments: I missed a discussion of the results. A discussion of the results is needed.
Response: Thank you for bringing this important issue to our attention. We have added Section 6.5 to discuss the advantages and limitations of ALSE.

八、总结

Yeah, that’s all.
感觉之后写论文的时候还要不停的翻看闵帆老师的博客，每一篇在论文写作中都大有用处，现在只是很浅显的学习了，在实践中才能出真知。

希望我也可以早点有实验成果，然后进入论文写作部分，陷入论文写作的烦恼！！

参考文献：Min Fan, How to write a boring paper in computer science?
友链 点击此处进入参考博客——闵帆老师的博客