I have been working on a research project and have submitted the paper to two different venues(not simultaneously), and both times I have gotten reviews which criticize the paper for a lack of novelty. What exactly makes an algorithm or system design novel? How can I judge the novelty of my solution?
4 Answers
The novelty of a scientific result is defined in terms of its relationship to previously published results. In order to perform such a comparison you need:
- to have a good idea of what related works have been published by the community, typically obtained by some combination of reading papers and attending conferences, and
- to compare your your system against the most similar prior systems and demonstrate its quantitative or qualitative superiority (e.g. "my system is three times faster" or "my system can process widgets AND frobs, and all previous systems could only do one or the other").
The "amount" of novelty is then how much better you are, relative to the interests of the community, and is very community dependent: a 2% improvement in the cost of manufacturing concrete is worth billions of dollars, while a 2% improvement in the speed of a personal computer program is generally unnoticeably tiny.
-
2The threshold for exceeding prior work also depends on the community. For example, a system that merely reproduces the results of other existing systems, even if it does so faster or simpler or more accurately, would not be considered novel in some fields, unless it also generalizes to a situation that no other model can handle.– David ZApr 22, 2015 at 12:58
Try to answer these questions and write your answers down in the "Related Work" section:
- What are the related algorithms?
- Is there any problem that they have not solve but your algorithm has?
- If yes, it's easy for you.
- If no, why is your algorithm different?
- Is it more efficient in terms of time or space?
If you have answered these questions, I think you are off to a good start to judge the novelty of you algorithm.
-
What if my algorithm is able to solve something that related work hasn't been able to solve. But, solving the same problem with their algorithm can be done by making some tweaks to it? In past reviews I have gotten feedback saying "one has to wonder if making X and U changes to Related work Z will not be sufficient enough to solve the problem"– RagHavenApr 22, 2015 at 8:45
-
@AndroidDev93 it depends on the "tweaks" you need to solve the problem. Try do write up why your algorithm is different. Better efficiency? Ease to use? Totally new way to solve the problem (greedy vs dynamic programming vs brute force vs random, just for an example)? What are the tweaks needed by the others to solve the special problem? When writing up the answers, it become clear to you if your algorithm is novel or not.– gefeiApr 22, 2015 at 8:54
-
5I would tweak the second point to "Is there any problem that your algorithm can solve that previous algorithms can't, even after minor modification?"– JeffEApr 22, 2015 at 12:41
It is very difficult to commensurate novelty. To make sure you score well at this topic you have to provide a good and fair comparison with what has been already developed in the field. You have to be well aware of the state of the art.
Similar to a mathematical theorem, novelty is demonstrated, and not implied. Don't expect the reviewer to do your work and compare your methods with other methods.
Also, try to see if the conclusions of any paper, that also contains a set of challenging problems, not solved yet, can be be addressed by your work. Also, try to get some position papers where the problem you solve is being acknowledged.
There are different orthogonal aspects you have to care about, novelty is one. Importance of the work is another, third is how to disseminate it easily to the public. History of science is full of people who made breakthrough discoveries and remained anonymous, while the scientist doing the mass dissemination actually got the credit.
I think you should know the answer more than the rest of us. It is novel if no one else has thought of it or implemented it. You must show this sufficiently in your literature survey (background section) to satisfy your reviewers.
As long as you can differentiate your work from others in the same field, then it is novel. I would think it is easy to show that an algorithm is novel, can't you test your algorithm against others that attempt to do the same thing, and test it using the same benchmark? Focus on computation expense and storage?
Feel free to send a rebuttal asking which paper(s) they think talk about the same thing, thereby making your paper not novel. Chances are, they have something to substantiate their claims.
-
6Differentiation isn't enough for novelty. Suppose that A has designed an algorithm that determines what colour a cat is, B has designed one that rates the cuteness of photos of black cats and C has designed one that rates orange cats. Combining these to give an algorithm that rates the cuteness of black and orange cats is clearly differentiated but not at all novel: it's a completely obvious application of existing knowledge. Apr 22, 2015 at 11:59