A Template and Suggestions for Writing Easy-to-Read Research Articles
aa r X i v : . [ c s . G L ] J u l A Template and Suggestions for WritingEasy-to-Read Research Articles
Tansu Alpcan
Abstract —The number of research papers written has beengrowing at least linearly – if not exponentially – in recent years. Inproportion, the amount of time a reader allocates per paper hasbeen decreasing. While an accessible paper will be appreciated bya large audience, hard-to-read papers may remain obscure for along time regardless of scientific merit. Unfortunately, there is stillinsufficient emphasis on good written and oral communicationskills in technical disciplines, especially in engineering.As an academic, I have realised over the years that I keeptelling my students the same things over and over again whenthey write papers, reports, presentations, and theses. This articlecontains some of those suggestions and serves as a limitedtemplate for organising research articles. I have adopted a verypractical and personal approach and don’t claim that this is aformal contribution to the scientific communication literature.However, I hope that this article will not only make my life abit easier but also help other graduate students and academicsupervisors. P RELIMINARIES
This paper serves as a template, contains suggestions, anddiscusses properties of well-written and well-organised papers.The templates for the specific sections are formatted as:
Template (Section) . Here is one way to organise this section: It should contain such and such.
Specific suggestions are emphasised using the following:
Suggestion (Section) . A few relevant tips and tricks: • use this approach when writing this section. The plain text parts of the paper discuss the best practices,present resources, and contain the author’s opinions.I. I
NTRODUCTION
I think almost all of my papers begin with a section titled“Introduction”. It makes a lot of sense to start the paper byintroducing the reader what the paper is about.Many authors fail to realise that the readers do not reallyknow anything about the author’s research. Specifically, stu-dents often assume that the reader has the necessary back-ground knowledge (including acronyms), appreciates the un-derlying problem they work on, and will fill in the blanks whenpresented the results. The reality is of course the opposite.Most of the time, the reader barely knows the topic and themethods, has no idea why the presented work is significantor useful, and cannot make sense of the results unless theimplications are explicitly discussed and clarified.It should be also crystal clear to any author that the firstpage of a paper is a “prime resource” in our time-constrained
Department of Electrical and Electronic Engineering, The University ofMelbourne, Australia. Email: tansu.alpcan @ unimelb.edu.au . ‘information age’. The first page is the face of the paper andas we all know, the first impressions matter a lot in humanpsychology. Unfortunately, many authors waste the half of thefirst page with banal generalities and filler sentences insteadof using it efficiently. Template 1 (Introduction) . A good way to organise theIntroduction Section is as follows: The author should first explain briefly the domain andthe main theme(s) of the paper. What is this paperabout? Most journals and conferences ask for a fewkeywords, which may provide a good clue. Next, the author should describe the motivation, intro-duce the main research question(s), and argue for theirsignificance. What is the question this paper tries toanswer and why should the reader care? The contributions presented in the paper and theirnovelty should be explained explicitly. What is donehere that was not done before in the literature? Is itthe model, theoretical contributions, simulations, exper-iments, or a combination of these? A brief comparisonto prominent existing works is absolutely necessary. Space permitting, it is a good idea to have a subsec-tion called “Contributions” and explicitly list the mainnovelty and contributions of the paper for improvedreadability. It is often customary to give an overview of the paperorganisation in a single paragraph, explicitly listing theremaining sections and what they are about.
Suggestion 1 (Introduction) . A few relevant tips and tricks forthe Introduction Section. • Ideally, the first three points in the template should fitto the first page. Hence, the reader gets an overview ofthe topic, the question(s) addressed, why these researchquestions are important, and what are the main contri-butions of the paper in comparison to the literature, allfrom the first page. • The introduction should focus on what the authors havedone and how (methodology). A common mistake isspending too many words in the Introduction on irrelevantgeneralities, background that does not further the mainstory, or the contributions of other works. • The introduction sets the tone for the upcoming discus-sions later in the result sections. If it is boring andconfusing, most technical readers jump to the modelsection quickly without having a good idea about whatproblem the paper aims to address and why.
I. L
ITERATURE R EVIEW
No scientific work exists in vacuum. Even the most inno-vative ideas have a connection to the existing literature. Infact, most of the scientific papers as of early 21st century arerather incremental in nature. The main job of the literaturereview section is to position the paper’s contributions withinthe literature. It also indicates the reader that the authors havedone their homework and checked what is already out therebefore claiming novelty of their contributions. If there is space,the literature review deserves its own section. If not, e.g. inconference papers, it can be embedded into the introductionsection.A common mistake the students do in this section is tosummarise each relevant paper they have read as part of theproject in couple of sentences without any organisation orgiving the reader any insights. Ideally, this section should beorganised just like a mini literature survey paper.
Template 2 (Literature Review) . Using the list of the contri-butions from the Introduction, start with a paragraph men-tioning the relevant background to each contribution item.This leads to a list of relevant topics that were previouslyexplored by others. Instead of a dry summary, the goal isto familiarise the reader with what was done before andhighlight the gaps in the literature. It may look like this: Starting paragraph: the topic has multiple intercon-nected aspects, A1-A3, which were explored before inthe literature... A1 was discussed early on in [1]. Another work [2]presented a novel xxx. However, yyy was not exploredbefore. Another important aspect is A2. The rich literature onA2 is summarised in the survey paper [3]. A3 was proposed by [4] and extended further by [5]. However, no paper combined A1 and A2 and extendedit to this new direction zzz to the best of our knowledge.
Suggestion 2 (Literature Review) . Good organisation is thekey to a good and readable literature review section. • Avoid listing one paper after another without any con-ceptual organisation. • Only mention the works that are directly relevant to thepaper themes. It is good to cite authoritative books andsurvey papers to save space and help the reader. • A good literature review should highlight the gaps, clarifythe contributions, and help the reader understand theposition of the paper in the grand scheme of things.
III. P
ROBLEM F ORMULATION
If there is sufficient space, it is good to dedicate a section toformulate the problems discussed in the paper. The title of thissection does not have to be “Problem Formulation.” Actually,it is often better to use a descriptive title from the specificproblem domain. This is the part where individual papers startto diverge from each other. If the problem is well-known,then this section can be embedded into the model section orthe preceding introduction section. If it is a new or complexproblem, then it may need to be explained to the reader in detail. The problem should also be properly motivated. Thissection should make the significance of the problem crystalclear to the reader.
Template 3 (Problem Formulation) . The overarching problemthis paper address is xxx, which can be decomposed into: How to achieve yyy as a preliminary step? What is the best method for zzz, so that together withyyy, xxx is addressed satisfactorily, i.e. the metric W ismaximised under constraints Y. Solving problem xxx has the following practical impli-cations: ...
Suggestion 3 (Problem Formulation) . This section is eitherembedded into one the others (model or introduction) or isclosely connected to them. • A brief summary of the problem should be on the firstpage of the paper (introduction) to prepare the reader tothis extended version. • If the problem arises within a specific model, then themodel section should come first or the problem canbe embedded into the end of the model section as asubsection. • Many technical papers do not bother to explain thereader why this problem is important and what solvingit will achieve. Pure mathematicians may not care howtheir art is used in practice but everyone else, e.g.engineers should tell the readers explicitly what solvingthis problem will bring.
IV. T HE M ODEL
Modelling is at the heart of modern science. Therefore, itis not surprising that most papers rely on some type of amodel. The model could be mathematical or algorithmic orsimulation-based or experimental, or a combination of these.The section title can be chosen accordingly, e.g. it could namedsimulation or experimental setup instead of the model. Again,descriptive titles are much more preferable over generic ones.The model used in the paper is the foundation of thepaper’s contributions. If the model is not explained welland the readers are left on a shaky foundation, they wouldnaturally neither understand, nor appreciate the hard work ofthe authors.
Template 4 (Model) . Most papers build upon a lot of back-ground knowledge and it is impossible to include all of it ina paper. Still, the author should try to provide the reader thematerial and pointers for completeness as much as the limitedspace permits. It is good to point out and give preliminary knowledgewith couple of sentences and good references. Thistells the knowledgeable readers where the authors comefrom and less knowledgeable ones clues about whatbackground is needed to understand the model. The model should be explained briefly and clearly. Itshould be supported by relevant references. • Mathematical: the relevant equations upon whichthe later sections build.
Algorithmic: describing the broad class of algo-rithms, which the contributed ones belong. • Simulation-based: the simulator, maybe why it waschosen with a sentence, the input data, high-levelsimulation setup. • Experimental: the experiment setup, the hardwareand software used, data... It is very important to highlight and discuss the underly-ing assumptions of the model (or limitations of the simu-lation/data/experiments). Hiding important assumptionsmade in the model is dishonest and bad practice. If desired, a subsection on the approach that is usedin the paper could be added to the end of the modelsection. They are different things, so a subsection headeris needed to differentiate. Alternatively, the approach canbe embedded to the (beginning of) result sections.
Suggestion 4 (Model) . The model section often does notpresent the paper’s contributions but lays the foundation onwhich they stand. • Many experienced readers jump to the model section veryquickly after a quick glance to the first page. For expertreaders, the model section is the face of the paper. • Given that most papers are published in specialisedjournals and conferences, it is natural to assume thatreaders have some background on the topics of the paper.The model section should be tailored to the readership.Something that is obvious to one set of readers couldbe mysterious to another depending on the researchcommunity. For example, the same contribution may needa totally different model (and background) section for dif-ferent venues. This is especially tricky in interdisciplinaryresearch. • The model is different from the approach used in thepaper. The approach subsection should smoothly bridgethe model and the result sections, regardless of beingplaced with the former or the latter. • Some researchers (unfortunately not too few) think thatby not clarifying the assumptions and limitations of theirmodels, they can oversell their results. I find this veryunethical and not so clever. Remember, as an author,one can fool many people some of the time and somepeople all the time, but not all the people all the time.Furthermore, all papers are archived for perpetuity. Howwill it look like ten or twenty years later?
V. T HE R ESULT S ECTIONS
Before presenting the beautiful and ground-breaking resultsof the paper, it may be a good idea to briefly explain theapproach used to obtain them. Why not include an approachsubsection or a paragraph or two, which clearly explain whatthe authors did to obtain the results and how?The results sections are the heart of the paper and containthe main contributions. Again, these contributions may bemathematical, algorithmic, simulation-based, experimental, ora mixture of these. The results should be presented overmultiple sections in a well-organised paper. It is hard to be prescriptive about the results sections due tothe diversity of research contributions. However, I would liketo make some suggestions based on past experience.
Suggestion 5 (Results) . Whether solving the important prob-lem described previously or introducing a novel methodol-ogy, the results sections present the main content. In fact,everything else in the paper is merely support material. Thepresentation should be organised carefully to communicatethe research contributions to the reader in an easy-to-followstructure. • Once a set of results are obtained, it is a good idea to takea step back and decide which results will be presentedin the paper. This requires a judgement on significance,which is not easy. – If there are too few results, then clearly additionalwork is needed; maybe further analysis of differentaspects of the problem or solution. – If there is too much material, hard choices need bemade on what to include within the allocated pagelimit, i.e which results are really important. Thesedays, one can always upload a longer version of thework to a repository such as Arxiv to provide moredetails. • Once the set of results is determined, that set needs tobe converted to an ordered set. The paper is necessarilyin a linear format so the author has to play the role ofstory teller to the reader, explaining the results one byone in an organised and logical way. This is easy if theauthor is clear on motivation and problem formulation.Most students (and some authors) struggle telling a storybecause they are confused about those two, i.e. why theydo what they do. In that case, the supervisors (or seniorco-authors) can give the much needed support. • Ideally, the paper should have a discussion section,subsection or paragraphs. Once the results are properlyconveyed to the reader, they need to be interpreted anddiscussed. Many technical papers fail to accomplish thistask. What do these results mean and what are theirimplications? These two points need to be explicitlyexplained and crystal clear. If the authors cannot achievethis maybe the authors themselves are confused and howcan they expect the reader not to be? • Visualisation is very important. We live in a visual centuryand a picture has always been worth a thousand words.Specific tips: – Graphs and bar graphs are better than tables, whichoften belong more to appendices. If a table has tobe used, the important cells and columns/rows shouldbe shaded/highlighted. – A graph should have a good title, legible axes labels,and a descriptive caption that stands on its own,e.g. “Algorithms A, B, C are compared under Yconditions for various Q. Curve A describing theoutput metric X of Algorithm A clearly outperformsCurve B of Algorithm B as quantity Q increases. etc.etc.”
Each graph should tell a story and highlight animportant result. Just because a graphical resultexists does not mean it is worth including in thepaper (and in that form). Many results can be toldverbally within the text if the contributions are trivialor as expected. Again, a link to the extended versioncan be provided to the reader.
VI. C
ONCLUSIONS
This section is very similar to the abstract and introductionbut also has a lot of advantages over them. Now that the readerhas seen all the results, one can provide a much more informedsummary that contains insights. The conclusion section shouldconvey the main points of the entire paper and the bottom lineof the work. This obviously requires the authors achievinggreat clarity in their understanding of their own work.
Template 5 (Conclusions) . This section may be thought like amini paper condensing the entire work, followed by the futuredirections. Background, motivation, problem addressed. The model, approach, and methods. A verbal summary of the main results, discussion, andimplications/importance. Future directions, e.g. what did not fit to this paper,could have done if time/resources permitted, or whatresults would have been nice to have based on theinsights of this paper.
VII. A
BSTRACT
It may look paradoxical but makes a lot of sense to writethe abstract chronologically at the very end. By that time,the authors hopefully have a very clear idea what the paper isabout and its messages. Here is a suggested template for theabstract:
Template 6 (Abstract) . Many abstracts in engineering articlesare deadly boring. In contrary, the abstract should excite andmotivate the readers to read the rest of the paper! One or two sentences on background and motivation. A sentence on the problem addressed. Couple of sentences on the model, approach, and meth-ods; their novelty in contrast to existing works. Highlights of exciting main results in two-three sen-tences. A closing sentence on the implications/importance of thecontributions/results. T IPS AND A DDITIONAL R ESOURCES
Technical Writing
English is today the de-facto language of science. Usingthis language properly is challenging for authors whose nativelanguage is not English (like myself). Clear writing requires,however, more than the expertise on the rules and mechanicsof a specific language. Achieving the clarity appropriate fora good scientific publication requires honesty, clear thinking, and maturity. The following three books may help in thisendeavour: J. Zobel, “Writing for Computer Science”. SpringerLondon, 2015.This book [1] has Computer Science in its title but shouldhelp engineers as well. It covers the fundamentals very well.Highly recommended. W. Strunk, E. White, and M. Kalman, “The Elements ofStyle”. A Penguin book : Reference, Penguin Books, 2007.This [2] is an absolute classic on the nuts and bolts ofwriting. A must read. J. Williams and G. Colomb, “Style: Toward Clarity andGrace”. Chicago guides to writing, editing, and publishing,University of Chicago Press, 1995.This [3] is the book to read once the author has a goodunderstanding of the basics. It is one of my favourites.
References
No scientific work exists in vacuum. The references shouldbe used appropriately • as pointers to the relevant background work. • as part of the literature review. The cited papers create thelandscape in which the paper lives. The editors often usethe authors of the related cited papers as a peer-reviewresource. This practical aspect should also be taken intoaccount! • to point directions and works that go beyond the scopeof the paper.The references should be appropriately formatted. Not fol-lowing the formatting rules, e.g. of IEEE or others as requiredby the publication venue, looks very sloppy and gives theimpression that the authors are not professionals. It is easyto find citations online but many repositories’ formats differfrom the required one, so it still requires some post-processing. Final Comments
Additional general tips and comments: • Always prefer explicit over implicit. Don’t expect thereader to come to conclusions and fill in the blanks;provide them clear material and your conclusions. Don’tworry, the readers will make up their own mind anywayonce they understand the material. The problem is whenthey have a foggy understanding of it. • Always use a spell checker! • Latex looks nicer than word processors but requires asignificant upfront investment. • Block diagrams, algorithms, and graphics that illustratethe concepts are very very good and should be usedliberally throughout the paper. One picture is worththousand words. • Scalable vector graphics (svg) or pdf, eps, should be pre-ferred over bitmap (png or jpg). Consider trying
Inkscape ,which is a good open source multi-platform program. • A scannable pape r is much more accessible than onethat is hard to understand. It roughly means: the readershould get a very good idea about the paper’s topicnd contributions just by looking at the first page, thegraphs/graphics with their captions, section titles, andconclusion. P RESENTATION C HECKLIST
Many scientists are quite bad in making presentations. Icannot count the number of boring, incomprehensible pre-sentations I have suffered personally in scientific conferences.Things are improving slowly with newer generations but still...This checklist is based on a mini course I took fromDeutsche Telekom in Germany years ago. Unfortunately, Iforgot the name of the instructor so cannot give him the credithe deserves for the nice job he had done.
Specify the cause of the presentation • Who proposed the presentation? Why? • Which significance does the presentation have from thepoint of view of xxx (in audience)?
Define the objectives and goals of your presentation • What are the factual and emotional objectives that I wantto achieve?
Analyse the expected audience • Which expectations, needs, wishes does my audiencehave? • How will the audience benefit from my presentation? • What are the main properties of the audience(background knowledge, education)? • Which prejudices may prevail?
Define the content • What are the key messages (Max. 3)? • What is the extend of background information needed? • Which examples enforce the effect? • What can be omitted?
Structure your presentation • In the introduction (catch them; 15%) • The main part (keep them; 75%) • Within the final part (convince them; 10%) • Pyramid Structure: does each slide have a message?
Define the storybook • Jokes, exaggerations, getting to the point: using pictures,quotations ... • Make I use of my own stories or ones from the experi-ences of my audience? • Do I use lively, active speech? • Do I present a change in presentation style every 10minutes?
Work on presentation style • How do I use my voice and body language ? • Are there enough short breaks in my presentation? (Theart of making a pause) • Am I personally convinced and glad to give my audiencesomething valuable? • Do I show my personal affection to the topic? • Do I have eye contact with the audience?He who presents is in the focus of the auditorium!Slides may be helpful but only as an auxiliary means! A
CKNOWLEDGEMENT
The author wishes to thank his current and past PhDstudents and postdocs for their feedback, comments, andsuggestions. R
EFERENCES[1] J. Zobel,
Writing for Computer Science .Springer London, 2015. [Online]. Available:https://books.google.com.au/books?id=LWCYBgAAQBAJ[2] W. Strunk, E. White, and M. Kalman,
The Elements of Style , ser. APenguin book : Reference. Penguin Books, 2007. [Online]. Available:https://books.google.com.au/books?id=sj5 wr6zIEcC[3] J. Williams and G. Colomb,