Some thoughts about writing a survey paperFebruary 25, 2008Regular research pap ers are a description of your own research. A surveypaper is a service to the scientific community. You are doing their research forthem. Instead of reading 20+ papers to understand what a scientific topic isabout, they just need to read your paper.Which subjects should you write a survey about: fields which are on theverge of maturity, but do not yet qualify for a book. If there are less than 10scientific papers in a field, do not write a survey. If all the 10 are from the sameauthor, do not write a survey. If there is already an exhaustive, recent survey,do not write anotherWhat should it go into a survey paper? The question needs to be asked inreverse: what do you want from a survey? How do you make the survey mostuseful to the readers?Introduction- A clear description of the field. What is it a subset of? What is thecurrent status?- Boilerplate is not useful → bla-bla-bla networks have seen a lot of interestin recent years...- short history: was there a seminal paper, research funding, special event,invention of an algorithm which spurred the development. Do not beafraid to anchor your domain in reality. 9/11 spurred a lot of researchdevelopment (and funding) in surveillance system. The introduction ofjava gave a new impetus to just-in-time compiler optimization research,and so on.- Which are the conferences, workshops, journals, special editions which arecarrying the papers related to the topic?TerminologyIntroduce the terminology of the field, describe what the various terms mean.What is very important is to map the terminological variations.For instance, in the sensor network domain, mobile sink, mobile agent, mo-bile data collectors usually means the same thing. In addition, some researchers1borrow terms like actuator, or invent specific new terms like “mole” for the samething.You need to clerify these things, so start by keeping a note of the variousterms while you are reading papers.Research challengesThe description of the various research challenges of the field. This is thehardest to write, because it is the part which is creative. You need to providean integral view on the research activity of the field.You should start noting out the various objective descriptions from variouspapers. But it is not enough to just put them together: you need to rewritethem in your own words. Partially for the reason of copyright, and second,because you need to write a good writeup, which covers all the papers not justone.One can call this step “reverse engineering a vision”. If your domain has a“vision” paper, that might help, but be cautious: you still can not borrow yourwriteup from the vision paper, if you really need to take a whole field.It is helpful to identify 3-4 main research directions, around which you willorganize your papers.Classification, slicing and dicing, taxonomySometimes it helps do introduce a new taxonomy that is classification schemein the field.The paper surveysAnd here comes where you survey your papers. This is how it goes:- Decide what are you going to tell about each paper. You need to alreadyread the paper in such a way that you know ahead what are you want totell about them.- For instance:- which one of the 3-4 big research directions they tell?- what mathematical techniques or algorithms they rely on? (eg. linearprogramming, genetic algorithms, neural network, hidden Markovmodels’ etc.).- is this a theory or application paper?- is it the continuation of another work? is it an improvement onanother work? (you might want to present them in order!!!).- do they use theoretical proofs? simulation? hardware testbed? reallife deployment?- which other technology they compare themselves with? In which wayare they better? Note: all the papers you will encounter are at least insome ways better than others. You need to identify the authors claim;higher performance (under certain assumptions)? higher robustness?lower computational complexity?2It would help if you would assemble your reading list completely first, andwhen you read the papers, you write down the answers to these question, asyou read them.About citations: strong recommendation that in your survey do indicate thecomes of the authors as well: Laurel and Hardy [2] did this, W. E. Coyote [3]did that.There are two reasons for this. One, it is politeness towards the authors,whose work you are surveying. Your debt to them is much greater than in acost of an original research paper. Second, when your reader read your surveypaper as a primer for a field, they are also interest in finding out, who are theresearchers active in the field. If some university or lab had a special leadershiprole, it is worth mentioning as well: Many early contributions in bla-bla-blanetworks come from the W.E.Loyote’s group at Hollywood Inst. of Tech Medialab.It almost always helps to creat a nice big table to summarize
View Full Document