My guide on how to write a related work section.
By Niklas Elmqvist, University of Maryland, College Park
The purpose of the literature review in an academic paper is to discuss the existing scientific literature in the topic of the work you are describing. While the introduction typically lists just a few important references, the literature review typically comes after the introduction and serves as a more thorough background for the paper. Writing a balanced, thoughtful, and exhaustive literature review is difficult and time-consuming. It requires the writer to be knowledgeable in the area they are writing about. Be prepared to spend a lot of time on your literature review. Below are some pointers on what to keep in mind when writing this section.
A key aspect of your literature review is its structure. As with any other part of your paper, the motto here should be that language clarity and reader comprehension is paramount. While the simplest approach is merely to list the related papers in the order you found and read them (i.e. as a laundry list), this is not helpful to the reader and also gives a sloppy impression.
Another idea is to use a chronological order based on publication year. This is perhaps a natural inclination, as progress over time is a central aspect of scientific discovery. A few times it might even be the right choice; for example, a few times I have taken a historical perspective on a topic (such as “The Rise of the GPU” in a paper on GPUs for visualization), in which case a chronological order makes sense. Most often, however, different threads of research belong together even if they happened several years or decades apart. Sometimes the exact casual order of past work is not important. In all of these cases, a chronology is not really useful as an organizing principle.
Instead of the laundry list or the chronological order, my preference is to organize papers in a logical order based on their topic. In my opinion, the most effective way and reader-friendly way to do this is to make these topics explicit by introducing subsections for each high-level topic; typically from three to five in total. This will help the reader to navigate in the section. Furthermore, the very best way to structure a literature review is to tell a story. Start from the beginning (not necessarily in terms of time, but in terms of foundations and basic ideas) and build from there. Sprinkle with references to support your argument. Make sure there is a structure apparent in your narrative; you should probably use the logical order with subsections, as discussed above. To reinforce the narrative aspect, implement the logical structure down to individual paragraphs, even sentences. Start each topic paragraph with a sentence to build the narrative and use bridging terms and segues to transition from one topic to another. Order the topics so that they support and build on each other. My old ScatterDice paper (InfoVis 2008) has a good example of such a narrative-based literature review.
Once you have the structure in place, what goes into the actual narrative? Basically, you need to describe the prior art in a detailed but concise manner. Being able to summarize whole research papers into one or two sentences (or sometimes just a few words) is a difficult but vital skill. This is where all your hard work learning how to read research paper comes into play! What is the essence of a paper, and what does it really add to the literature? Reviewers want to learn from your view of the field and appreciate succinct and accurate descriptions of existing work (particularly if it is their own!).
At the same time, fairness is important when characterizing other people’s work. You need to be able to stand behind any statement you make in your related work (and, indeed, in any part of your paper). It is perfectly fine to criticize existing work—after all, if previous work did not have flaws there would be no need for your paper—but you should write your criticism in such a way that you could tell it to the face of the author of that paper. Remember that reviewers are supposed to be experts in the field of the paper they are reviewing, which means it is very likely that at least one of your reviewers will be an author of a paper you are citing. In fact, an explicit strategy of many PC members is to select reviewers from the list of references in the paper. If you make their work look bad or attack them unfairly, they may punish you for it.
The literature survey explains the current state of the art in the field, but if your work does not contribute anything new, it is by definition not needed. For this reason, you need to explain how your work differs from (and hopefully improves on) what is already out there. This is not necessary for all papers you reference; some of them are clearly very different from your work. However, it is particularly important for the papers that are close to your work. There’s a chicken-and-egg problem here: explaining the difference between a related piece of work and yours is difficult before you have explained your own work in detail, which happens later in the paper! Typically you can solve this by explaining your contribution in broad strokes, but it is clear that the more similar your work is to existing work, the more details you must provide. A more advanced way to bypass this problem is to try and define the problem (i.e. not your solution) in a couple of orthogonal and adequate dimensions, and then use these dimensions to classify the existing literature. In other words, by finding these dimensions, you are creating a mini-taxonomy that you can use to showcase the gaps in the state of the art. If you do it well, it will be clear to the reader that there is more work to be done in this area. In fact, the taxonomy is almost a contribution in itself. Some examples of this include our InsightsFeed paper (EuroVis 2017), multiple line graph perception (InfoVis 2010), Munin (TVCG 2014), as well as our old Melange work (CHI 2008).
Beyond these guidelines on the structure and content of a typical literature review, here are some additional considerations you may want to take into account when writing it:
- Literature reviews are important! You should understand that a related work is not just some arbitrary hoop that every paper has to jump through but an intrinsic part of an academic paper. Science is founded on science, and a good command of the literature is a hallmark of a good scientist. More importantly, people appreciate a good literature survey, and you will win the trust of your audience if you are able to adroitly position your work in the greater scheme of things.
- Be mindful of plagiarism! Writing about other people’s work can mean skating a thin line between original writing and plagiarism. If you are paraphrasing text from another paper, it is fine to just add a reference, but if you find yourself quoting it outright, you need to add quotation marks and list it explicitly as a quote (with page numbers)! Remember that copying text from someone else’s literature review is plagiarism, even if it is text about work these authors did not perform themselves. This cannot be stressed enough. Plagiarism kills careers. Do not let it happen to you (or your co-authors).
- Ditto self-plagiarism: Literature reviews are often formulaic, so you may be tempted to merely copy text from one of your previous papers. However, this is a form of self-plagiarism. For one thing, your paper may be flagged as being plagiarized by a double-blind reviewer who can’t know your identity. The best solution is to retain the actual references but rewrite the narrative itself.
- At the end of the paper? Another solution to the aforementioned chicken-and-egg problem is to put the related work at the very end of the paper. This is uncommon, but for some papers it works well—particularly for those where a lengthy discussion of the problem you are trying to solve is not strictly necessary. In this situation, it is quite easy to compare the existing work with your technique, because the reader will be intimately familiar with your technique by the time he gets to the related work. I rarely use this method myself.
- Literature takeaways: Your literature review is supposed to inform your current work, and one way to make this explicit is to include the actual takeaways in the paper itself. For example, in our recent Vistrates paper, we include a subsection titled “Summary of Related Work” at the end of the literature survey, and in our ATOM paper, we do the same in a subsection called “Contributions”. Taking this a step further, in our Branch-Explore-Merge paper (ITS 2012), each subsection is followed by a small bullet list of explicit insights that we base our current work on.
While literature reviews follow a relatively strict format, there still is some latitude for experimentation. Some of the tips above may help you find an effective organization that will help the reader orient themselves in the literature.