Writing a Related Work Section

This post outlines my approach to writing a Related Work section. Some of this is subjective, take it with a grain of salt.

Finding Related Papers

Writing the Related Work Section

Tips

1. Start writing this section early.
2. Your paper should be related to, but not identical to other work in your topic area.
3. Think about this section from the perspective of a reviewer.

The reviewer is skeptical. There is a lot of published research out there. It's easier to duplicate an idea than to contribute something novel. The reviewer is frugal with their cognitive effort. In assessing the novelty of your paper, they will first make assessments requiring little time and energy. At the same time, they love the topic area, and will be excited to see a genuinely new and interesting contribution.

4. Understand where your paper sits on multiple levels.
   1. What themes are you working on? What’s your general topic area?
   2. What are your research questions or hypotheses?
   3. What methods do you use to answer the research questions?
   4. What results did you obtain by applying these methods?

Mental Model

This diagram shows four levels that readers might use to “pigeonhole” your paper. Readers need to answer “what is this paper about?” There are different ways to answer this question, some low effort (“It’s about large language models”), and some high effort (“this paper assess how large language models think about sequential move games using interpretability techniques, finding that game state representation significatly affects model performance”).

In an effort to conserve mental energy, readers will first attempt to answer “What is your paper about?” at the level of themes and keywords, then research questions, then methodology, then results. Readers can make comparisons between your paper and the existing literature at any and all of these levels. The Related Work section can (and should, in my opinion) fast-track this process for the reader.

The below diagram illustrates this thought process. The lefthand side shows levels of comparison. The righthand side shows how comparison outcomes affect the reader’s assessment of your paper’s novelty.

By writing your related work section with this thought process in mind, you can write in a way that benefits your reader.

Examples

Here are some examples of the kinds of statements you might use to clarify your relationship to existing work at different levels.

These examples were written as part of a conversation with Claude 3.5, with what I would consider moderate steering. I have read the examples, and think they generally show what they are intended to show.

Relationship Type	Example Statements
Different Context Same methods, new domain	"While Smith et al. [1] applied similar visualization techniques to financial data, we extend these approaches to educational analytics..." "The optimization method of Jones [2] was developed for manufacturing processes, but we show it can be effectively adapted to healthcare scheduling..." "Although this algorithmic approach has been well-studied in desktop environments [3,4], our work is the first to explore its application in mobile contexts."
Different Goals Similar domain, new objectives	"Previous work in this domain focused on maximizing accuracy [5], whereas we prioritize computational efficiency..." "While existing studies examined user engagement [6,7], our research investigates the long-term retention implications..." "Prior research analyzed these patterns to predict behavior [8]; we instead use them to understand underlying motivations..."
Novel Approach Same problem, new method	"Unlike previous solutions that rely on supervised learning [9], our approach leverages unsupervised techniques..." "In contrast to the statistical methods commonly used in this domain [10,11], we introduce a geometric approach that..." "Where existing work has approached this using rule-based systems [12], we demonstrate that deep learning can..."
Different Findings Similar setup, new results	"Contrary to Wilson's findings [13], our larger dataset reveals that this correlation weakens at scale..." "While previous studies found a linear relationship [14], our more granular analysis shows non-linear effects..." "Our results challenge the widely-accepted model [15], showing that under certain conditions..."
Building Upon Similarities When acknowledging foundations	"Building upon the foundational framework established by Brown [16], we extend..." "Our methodology follows the validated approach of Lee et al. [17], adding refinements for..." "We adopt the core metrics proposed by Garcia [18], complementing them with new measures for..."

Antipatterns

1. Considering work from Theme 1 and Theme 2, but not Theme 1 $cap$ Theme 2.

More reading

-Literature Review for Academic Outsiders - Related Work can be thought of as a small version of a Literature Review, much of the same advice applies. - There are more linked resources inside of this linked article.