How to get your thesis topic started

Students often struggle to get their thesis topic started properly, because they simply don't know how to start at all. In the end students may find that they have missed and left out major areas which would be in their topic's focus - which makes their thesis incomplete then. Correcting these issues regularly leaves students with a very huge amount of extra work. To address this issues this small guide has been created, which gives a short overview of how to get your thesis started, and - as a part of the task - how to effectively read papers.

Major steps for getting your thesis topic started

  1. Learn how to effectively skim through/read papers
  2. Get an thorough overview of research done in your topic. You need to develop a "feeling" of what is already out there, and it's important to ensure to *not* miss any important sub-areas inside your research area.
  3. When reading papers, keep a well structured reference list from the start (e.g. JabRef) of all literature you find interesting.
  4. When having an overview and feeling of what is already there: discuss with your supervisor a) what is there and b) what to do next given this knowledge.
  5. Ensure you have finished all that before doing the major implementation of your thesis and writing the thesis itself.

How to get an overview of research done in your topic's area, or: how to search for papers

  • You need to search for publications (see scientific search engines below). Therefore, think of phrases that represent what you search for and will be contained in publication titles of the research area. For the start, they can be very easy phrases, like for face detection on mobile devices: "mobile face detection survey" or "mobile face detection review".
  • For first searchings, combine your phrases with the words "survey", "review" etc. as above. This way you will find publications that themselves sum up research in a particular research are up to the publication time. Such survey papers do this by listing/citing most important previous publications themselves (like with textbooks - for books you can use search terms containing e.g. "introduction" or "book"). Through these publications you will get a feeling for how researchers *call* their approaches - which will be very useful for future literature search.
  • Starting with the last point: for every publication you read and find interesting: keep a complete reference of the publication in your reference program (e.g. JabRef). Complete means including the abstract and publication file (if possible). Also write a very short "review" of the paper yourself to later immediately remember what the paper was about (e.g. "gives an overview of X with comparing approaches X1, X2 and X3" or "tries to solve problem X by using method Y. find out that [...]"). E.g. bibtex provides a "review" tag for this matter.
  • When read about something interesting in a publication that itself is cited there: jump to the reference list and grab the reference for further reading.
  • When you find a paper X interesting, check if there exist newer papers that cite paper X. Those papers might be interesting to you as the might have done further progress in research, based on paper X.

Where to search for papers

Use scientific search engines ("online libraries" for our sciences). Some libraries related to our sciences are listed below. Inside our university network you can download complete publications for most of those libraries (which is not possible outside the network - there you would have to pay for the access. Details on:

  • (very good as starting point)

How to effectively read a paper

Papers are almost always structured somehow similar to the following:

  • Title/Authors (most important: what is the paper in one sentence?)
  • Abstract (second most important: what is done in the paper?)
  • Introduction (what is the problem? - in detail)
  • Related work (other research the authors are aware of when writing their publication. Good for further reading)
  • Method / Approach (description and details how they address the problem)
  • Implementation + Evaluation (practically proving and measuring the method to see "how good it is")
  • Conclusion (third most important: what was done in the paper?)
  • References (corresponds to related work)

From the title decide if the paper seems related to your topic. If yes or if you are not sure about it: read the paper's abstract. Abstracts are short, therefore fast to read - and as for papers, abstracts are often structured similarly:

  • What is the problem the authors address?
  • Why is this problem important / why is it still unsolved / what are it's difficulties?
  • Which approach do the authors propose? What do the authors therefore contribute to the research field?
  • What is the method the authors use in their approach?
  • How did they evaluate their problem?
  • What are the evaluation results? What can be concluded from these results for the research field?

From a first skim of the abstract you should be able to decide if the paper targets your research area. If yes: read the abstract at least so carefully that you can roughly answer yourself the questions above. If you find the paper still to be interesting for you, skim through the conclusion. The conclusion essentially tells you what the authors did in the paper (so it's similar to the abstract, also in length - but written from the point of view that you *already read* the paper). When reading the conclusion you can re-check that your assumptions about the paper's content were correct. If the paper seems still interesting you can now start "reading" the paper. Skim through the introduction - if you already know about the problem the authors address in the publication in detail. If you do not yet, read it carefully to understand the details of the problem the authors want to address. Read through the related work to get an overview of the literature the authors were aware of when writing their publication. This is a good point to look out for other papers that are directly correlated to this paper for further reading. The sections from method to evaluation go into more detail on the approach the authors used to address the stated problem. Here, a "good paper" is written so that readers can fully reimplement and reevaluate the method - as it states all necessary details. Therefore, it is often sufficient if you dont' understand everything in full detail (except you want to write about this exact approach, reimplement it etc.). But read the method at least so carefully that you can answer somebody else the question: how did the authors address the problem/what is their method? What were findings for their approach within the evaluation?

Write a very short review - in your own words

After finishing reading the paper state very short in your own words what the paper is about in your reference as a review (as stated, bibtex provides a "review" tag for this). This review is mainly for you to immediately remember what the paper is about/how they did what the did - without needing to re-read the whole paper.

Further reading - starting at this publication

Now, after having finished a paper X you found it interesting: if you would like to read futher, related publications (besides reading the related work cited in the paper) you can search for papers that cite paper X themselves. Those papers might be interesting as they might have further progressed in research based on the findings of paper X. To do so search for your paper on scientific search engines and look out for tabs like "cited by" or "active bibliography".

Two more things

First: there exist multiple types of papers.
Short paper: e.g. <=4 pages: a shorter version of a paper. Does not go in too much details, therefore often has less references or evaluation. Often the purpose is stating an idea.
Paper: e.g 7-10 pages: "normal" paper, basically as described above.
Article: longer and more comprehensive version of a paper. Describes the problem, related work and method in even more detail, does a extensive evaluation. Articles are often extended version of "normal" papers, so you might find a paper and an article describing the same thing - with the article being the more comprehensive publication.

Second: how to estimate the quality of a paper you are reading?
Be careful in case the paper seems completely unstructured or major parts are missing (e.g. there is basically no related work stated in the paper or the evaluation of the method is completely missing). It's often a hint of reduced quality if a paper contains many typos, formatting errors or if it has (as a non short paper) only a couple of references (not counting internet references). Usually normal papers have somewhere in the range of 10-30 references.