Research-III: Types of Research

Research can be classified in different ways, depending on different yardsticks. One is to look at the methodology adopted for the research work. And another is dependent on the nature of problem being solved. We briefly look at both these here. Given my familiarity with computer science related topics, I will focus on this domain, mostly; though, I guess, most of my observations will hold good for other disciplines also.

Typically research involves posing a problem, analysing it to understand its various aspects, looking at literature to see what related aspects have been attended to and how, making some hypothesis towards solving the problem, and checking if the hypothesis is true. At each step, one can improvise and refine the process, depending on the domain and nature of problem.

Let us say, for example, we want to look at learning natural language. We find that the existing language learning systems are not very effective – there are very few users, completion rate is low, and may be, the quality of learning is not good enough. You can look at how to improve this. An analysis of the existing work may throw up a number of possible suspects. The failure could be lack of interaction; it could be inappropriate pedagogy; it could be the quality of content (hard to follow, too fast paced, etc); and so on. You can now pick one or more of these, and look at how to enhance this aspect. A literature study would help you to see what kind of things people do in this area – what are generally good pedagogy models, for example? You can also think/meditate/discuss with friends/etc and come up with some innovative approach. You would need to argue why you think this is a better approach. And now you need to see if this actually performs better. You may build a system and get a good cross section of users to try it out, in comparison one of the existing systems, collect performance parameters and analyse.

This also happens to be an example of experimental research. You conduct experimental studies – empirical – to show something works or not. The problem with this is that, this does not prove your hypothesis. But with a choice of large enough test data set, carefully selected based on an analysis of profiles and relevant paramenters, and careful formulation of the experiment, this is often a useful result. In many areas, formal results are too hard.

Yes, the alternative is a formal method. In some domains this is the recommended approach – eg, analysis of algorithm, proving complexity, proving equivalence of things (eg. A set of algorithms or formulations), etc. One uses mathematical models and results to do these. Deriving lower bound for the worst case complexity of a class of algorithms also belongs here. Empirical studies do not suffice at all.

Another common kind of work is formulation of a framework, protocol, etc. These problems need an approach that is dictated by the specific problem. The challenge here is to ensure that the proposed framework/protocol is not too general, nor too specific; substantiating this is a challenge in such problems.

Another category of research is based on surveys. Usually, you are looking for some patterns across a population. You may analyse software projects across companies, company performance, employee behaviour, web user behaviour, medical records, supermarket purchases, etc. The field of data analysis and visualisation is very much concerned with such requirement. You need to ensure good quality and quantity of data. Quality includes completeness (not too many records have missing values), correctness (values are mostly correct), coverage (adequate data points covering relevant sub-populations), and so on. The analysis of the survey results can be fully manual (when you can also include qualitative analysis), or automated. Looking for patterns without any hypothesis is like looking for a needle in a haystack, purely driven by luck.

If you look at the survey papers in ACM Computing surveys, you can see yet another type of research. A survey paper reviews most papers in a relevant field, and attempts to extract broad directions in that field and general trends. This may appear to be very easy, as some of the literature survey sections in thesis works do – just list the papers with a brief summary, and some general remarks. This methodology would result in quite a poor survey. The challenge is to identify connections among the different papers, which may not be reported by anyone, and based on them, group the work into interesting segments. This grouping is often a key insight into the domain – and that is what makes these survey papers valuable. Such a grouping requires good research experience in the domain, a thorough understanding of the domain, and wide reading.

One can also look at comparative studies as useful research, when done carefully. Comparing N algorithms should not be just reporting run time over R datasets, for example. The items to be compared need to be carefully articulated to reflect the basis of comparison. For example, an algorithm is not the same as a particular implementation of the algorithm. So, you need to know if you are comparing an algorithm or an implementation of it. The comparison can be theoretical (as in the case of equivalence relations) or empirical (run time performance, for example). Choice of background assumptions can be very crucial in these cases – every algorithm, for example, has its strong points and weak points. Recall that the simple inefficient bubblesort is often faster than the efficient quicksort, when the data set is small or nearly sorted. So, datasets or experiments need to be carefully formulated, to bring out the best and worst in each. And that is the challenge in ensuring good quality in these kinds of works.

This is not as organised and complete as I wanted. I need more thinking/shaping here. Meanwhile, it is a start. And share your thoughts and other feedback.


2 Responses to “Research-III: Types of Research”

  1. Sandip Saha Says:

    Yup in Financial,Marketing and HR domain also we follow the same procedure.We do literature study then carefully form those hypothesis after that conduct a market survey and collect data(collection of data can be in various way such as setting questioners , already existing data available in the market etc. ). Then we perform various statistical test to accept those hypothesis based on result . We also do comparative analysis and see other qualitative aspect .Finally after viewing the problem in 360 degree angle we come to conclusion.

    • thelittlesasi Says:

      As I said, what I am saying is not restricted to any given domain… because the broad procedure is similar. But the details and the ‘research’ element can vary a lot. Often the survey based “research” are quite shallow questions. There is also a lot of variations in what people mean by ‘literature’, ‘analysis’, etc. But, yes, the overall principle applies.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: