What if my research does not give the expected results?

Question

I'm working on my Computer Science thesis. I have a two solution proposals until this moment, and I've been wondering about the following:

Sometimes you know that if you implement it correctly, it will work (eg. when building a website or a desktop app for some trivial purpose, etc), but some other times you try to solve your problem with some specific solution proposal. A few examples:

• I will do some preprocessing, I'll represent the documents that way and use this metric, and then I'll apply this clustering algorithm. Theoretically, it should work.
• I will use this computer vision technique for breaking the captchas, and then... In theory, it should work.

but when you implement it, you don't get the expected results (eg. bad text classification in the corpus you're using, low-rate successfuly captchas solved, etc.)

In those cases, is it a valid work? I mean, is it good to publish your work, saying that your solution proposal doesn't seem to be good when solving some specific problem?

Answer 1

I'm not sure what you mean by "theoretically it should work". The examples you refer to are cases where you're positing some kind of model for the data. That the experiments failed suggests that the implied model is wrong. The question then is: what is a good model, and what went wrong with the model you tried.

It's that kind of investigation that will set you on the path to a paper.

Answer 2

In those cases, is it a valid work? I mean, is it good to publish your work, saying that your solution proposal doesn't seem to be good when solving some specific problem?

At least two advice:

1. Don't worry
2. Don't succumb to the urge to publish immediately

There is always a venue to report about your work, however little significance it bears. Workshops collocated with conferences are places where you can honestly report your work, including dead ends.

I recognize a more important thing going on here. You embarked on a research, from what you wrote, it seems you are convinced that the problem is sound and worth pursuing, but so far performed an experiment in the field only to find out you reinvented a wheel, or the thing is not good enough. One way of seeing what happened is that it is a kind of a failure. That's what you seem to suggest. Another optics applicable on your situation is that you embarked on something worthwhile, but since it is not an easy problem, the first approach did not yield a result. There are two observation to make here:

1. your problem is probably a good one. Easy problems, or non-problems yield a result usually very quickly (any approach is good enough);
2. you are on the right track to discover something. Eventually. Realize that the path from nowhere to the bleeding edge of human knowledge (state of the art) is never an easy one. You started and did not arrive to the edge yet. Expect few more experiments of "reinventing a wheel" flavour. After a while you'll get to the boundary and since you walked the path the hard way, you will have all your weapons sharp and enough insight to push the boundary. That's the positive message about your situation.

Answer 3

Not every solution method is a valid approach for solving every problem. When you believe you have a tool that is universal, then you start running into the "hammer complex": when all you have is a hammer, everything starts looking like a nail, even if it's actually a screw.

What I mean by this is that this is a reasonable finding. Acknowledging the benefits and disadvantages of your strategy is an important part of open science. It makes it easier for others to follow on, adapt, and adopt your work. If you declare a problem "solved," it makes it much more difficult for others (including yourself) to continue to work in the same "problem space."

Answer 4

Getting unexpected results is not problematic, it happens more times than you think. Note that with these test results you have achieved something, even if different from what you expected, but it is something that may be relevant to your research.