-
Why do you need a research question?
The function is to structure your thesis, your work, research and everything you do during the thesis period. It is also meant to make sure your reader (examiner) always knows why she is reading something. The main research question with sub-questions are the backbone, or even the whole skeleton, of your thesis. They provide the basis for a smooth reading experience, a clearly built structure and a convincing argument. -
A tree-like structure of (sub) research questions can work
well:
- The standard setup is: depth 3, branching factor 3, numbered RQs.
- So: RQ, SRQ1, SRQ2, SRQ3, SRQ1.1, SRQ1.2, ....
- The root is a general question making clear what you will do. Make sure that it is not a yes/no question. It can be helpful to start with “to what extent” and to make sure that it is comparative with respect to a baseline.
- Your sub RQs provide the evidence with which you will answer the root RQ in your conclusion. Make sure that it is a question that is answered by what you did.
- Your leaf sub RQs are super concrete. Readers should be able to predict the shape of the answer (e.g., a PR curve, a set of F1 scores for different systems plus indication if differences with a baseline are significant, etc.).
- All RQs should be easy to read and understand. The higher in the tree, the more they are understandable by non-experts. For every paragraph, every subsection, every section in your thesis, it should be clear to which (S)RQ it is connected. Everything you do in the three months need to be connected to a (S)RQ. And vice versa, you need to answer all your (S) RQs.
-
What does not count as a research question?
- All questions for which others have provided the answer already. All things you can find in the literature can thus never be the answer to a RQ.
- To find the answers to your RQs, you must do research! This means that you work with data, do experiments, code, think, play, experiment.
- Avoid formulating questions that can be answered trivially, like "Is it possible...." (yes, everything is possible, well almost everything).
-
You will find two examples of tricky cases below:
- What is the optimal... That is a pretty hard question to answer if it is not constrained very heavily.
- Which AI/ML/... method for my XYZ problem leads to higher profits for my company? Can you actually test that? Which test design is needed for it? Will your organization/company allow you to do that? Maybe it was the motivation to start the project for your company, but a motivation is usually not an RQ. Make sure that the question is scientifically relevant.
-
For Data Science projects, it is helpful to combine the
Explorative Data Analysis (EDA) with your RQs and ask yourself
the following questions:
- Do I have the data, the features and the gold standard train and test sets to answer each of my RQs and SRQs?
- Is it crystal clear what data I need for each of my SSRQs? What does the quality and the "richness/content" of the data need to be?