Hello, Prof. Philippe Fournier-Viger.

I'm a master's degree student and for final project , I'm researching on "Top-k high utility itemsets mining". During this period of study, I had studies in the field of data mining and implemented in C#, such as Two-phase, Hui-miner, La prune, Efim, TKEH, kHMC ,...

Now I'm looking for a new idea for top-k mining. I'm having trouble at this stage of the research. Can you discuss an innovation?

For example : Provide efficient pruning data or combine data structure with available pruning methods to improve runtime and memory or top-k in negative datasets

Regards

Hi,

I think that the fact that you have implemented many algorithms shows that you have good coding skills and understand well the algorithms. This is a good first step.

Now, I understand that finding an improvement is not easy. On the topic of top-k high utility itemset mining, there are two main possible approach:
(1) make a faster or more memory efficient algorithm,
(2) or try to integrate some new ideas in the top-k mining problem to make it more interesting.

About (1), in my opinion, it is not so interesting, because there have been several algorithms on this topic in recent year. Thus, maybe it is not so easy to come up with a new idea for improving the performance. Sometimes researchers will just reuse past ideas from other top-k problems but that is not very original.

To make a good paper, you could consider approach (2). In that case, you need to add some new element or idea to the top-k problem. For example, there are papers on periodic high utility itemset mining or average utility itemset mining. How about combining this with top-k? If it is easy to combine, then it is not interesting... But if combining such ideas with top-k raises some new challenges, then it will make a good paper. Above, I just mention periodic pattern and average utility as example, but I do not necessarily think that it would be a good idea to do that. Also, another key point is to think about the potential applications. If you design a new problem or add some new idea to the problem, you need to show or at least convince that it is useful for some scenarios.

So overall, I think top-k high utility itemset mining is a good topic, but if I were you, I would try to add something to the problem to make it more special. For example, top-k closed or maximal high utility itemsets, may be challenging and useful, and perhaps no one has done it. If it raises some new challenges, it could be an interesting topic.

But all of this requires to find some time... Finding a good idea is not always easy. Currently, I am also working on finding ideas for my students, and it takes quite a bit of time ;-) But it is very important to make good papers.



Edited 1 time(s). Last edit at 07/04/2019 08:05AM by webmasterphilfv.

"If it is easy to combine, then it is not interesting..."

That's exactly the same. You are the best professor. I am not fluent in English, And I can not really convey my sense!

Thank you very much for your complete description. Now I am studying several other papers. I hope to give you some good news from this topic

You are welcome. Hope you find some good idea soon for your research.