VU computer science students are required to perform a literature study course (X_405111) in an area of their choosing. This work is an important exercise in building a skill set for understanding an area of research by searching papers, summarizing them, connecting the dots between different papers, building an overarching narrative, and writing scientific documents. However, due to the open nature of the work, it is challenging to find a way to start the work. I often find myself giving the same advice to multiple students. Here is my attempt to summarize these points.

• It is advised to pick up an area of work which will closely match with your thesis work. It should not be the exact topic, but something border and more encompassing. For example, if you plan to write a thesis on implementing a new file system, then the survey could be about recent advancements in the storage technologies.

• In our initial meetings I will give a couple of seed papers and a potential list of conferences that you should search through.

• Come up with a paragraph of motivation for the survey. Why does this survey make sense now? What kind of “transferable” knowledge will this survey generate? What is the research question that you are attempting to answer?

• I am often asked how many papers or pages are enough. There is no definite answer to this question. I hope that you are not doing a survey with me just to fill up pages. The scope of the survey depends on the research activities on the field. For a well established area, you might encounter be 100s of papers whereas for a relatively new field of search there might only be only 10 papers.

• The next question is how detailed the summaries should be. The nature of your summaries depends how you are writing or presenting your survey. You need to extract the supporting details from a paper to substantiate your survey motivation and research question.

• A good survey will summarize the state of the art in a field in a presentable and consumable manner while working towards answering the research question. For this, I would strongly recommend thinking about what kind of visual aids (tables, figures, graphs) can you use to enhance the presentation of your survey.

• A good survey also identifies potential research directions or research opportunities in an area. This will also open a smooth transition to your actual thesis work. Do not just write isolated summaries of papers.

• As a work plan I strongly recommend to start early. Consider the following work plan to make progress:
  • Read the seed papers, mark their related work and citations.
  • Start backward (from their related work) and forward (from citations of the seed papers) searches to identify further key papers. You can use Google Scholar for this.
  • Quickly read the introduction and related work sections of newly found papers to get an overall idea of their work.
  • Try to build a structure of the survey - what heading, what subsection heading, what group of papers attempt to solve a similar theme of problem. This step may not be obvious, hence it may require a couple of attempts. It is a continuous process and the more papers you will read, hopefully the clearer the structure of the survey becomes.
  • In my experience, at some point in this process you will stop making structural changes (e.g, new subsections) to your survey and the “most” of the existing work will easily fall into the subsections that you already have.
  • Come up with a skeleton survey draft and try to attempt to place the papers in various subsections. A single paper might have contributed to two or more subsections. We will have a discussion at this point. You should present your ideas and vision of the survey.
  • After this, you can start reading the papers more carefully and synthesize pertinent details for your survey and start filling out the skeleton draft. Perhaps, in this process you have to make another adjustment to your survey.

• A good survey should be comprehensive. This statement does not mean that you must include all the papers published under the sun since the Genesis. In my opinion, it is very important for you to identify broad “classes” of work in the area of your survey. A missing paper can be added later, but a missing “class of work”, which might have altered your survey structure, theme or conclusions, is not good. There is a fine balance between breadth and depth in a survey.

• The literature study is supposed to cover a relatively broad area of work. Hence, start early and make a rough timeline plan. A good survey takes time, not just for writing but also to think and giving time for paper details to settle down. There are no shortcuts to this process.

• A good survey involves lots of creative work. The same set of papers can be presented in multiple ways. Hence, take this opportunity to make your own unique survey.

• For completeness you should also include (1) your methodology and tools used; (2) referecnes to any other similar or related surveys that are available; (3) limitations of your work; (4) an outlook section in the end where you should summarize your impressions of the field, and how do you see the overall picture going forward.

• For practical work, I recommend to use the USENIX LaTeX template for the survey (available here: https://www.usenix.org/conferences/author-resources/paper-templates) and keep a progress log that you should share with me.

Lastly, here are the evaluation criteria for your survey:

• Quality of study design
• Rationale for literature selection
• Quality of results: scientific quality of the analysis, discussion of the findings
• Quality of writing: style, clarity, organization, grammar and orthography
• Correctness and completeness of references and citations
• Final presentation: final presentation (narrative style, balance between focus and coverage of contents, answering questions)
• Independence in the work

Computer Science is a fast moving field. Hence, good surveys are the closest reading material we have for the textbooks. A good survey from you will be publishable and perhaps used in courses.

Having said all of this, I am curious to hear your opinion about the process. Feel free to discuss with me what works and does not work for you, or if you have a different way of approaching survey work.

Good luck with the work!

Further reading:

1. ACM regularly publishes a series of high quality surveys that you can check for their motivations, structures, writing styles: https://dl.acm.org/journal/csur.

2. Also check out the “Useful links” section for scientific writing, and dos and don’ts in writing.

---

Jekyll theme inspired by researcher