-2

I read a paper about Rumor detection and they used BERT as an unsupervised language representation, fine-tuning it using a small dataset, and combining it with a supervised learning model to provide an enriched text representation of the content of the rumor, Now I achieved the same accuracy using sentence-Bert, what are the differences between these two models and I want to know if their model is slower than mine or not, they did not mention the speed on their paper. they also have another paper with lower accuracy can I mention the paper that has lower accuracy to compare it with mine?

2
  • What do you mean by "what are the differences between these two models" in the sentence? Could you please rephrase your question, it is currently hard to understand.
    – GoodDeeds
    Commented Oct 18, 2021 at 21:14
  • 3
    This question is about the content of machine learning research, and hence beyond the scope of this community. Perhaps try at crossvalidated.SE
    – user116675
    Commented Oct 18, 2021 at 21:45

1 Answer 1

-1

welcome to academia!

Now, you are the expert in the field and probably more in a position to answer "what are the differences between these two models" part than anyone here ;p

The only relevant to academia.SE bit needs addressing though. You are aware of their paper showing an accuracy same as yours and choose to cite a different one, with worse accuracy, to present your own results in a favorable light and possibly invent a reason for your work to have merit and be published. Nope-nope-nope.

It is a serious lack of integrity, you should never ever do such a thing (being a reviewer, I'd do you a favor and assume incompetence rather than malice but it's still not what you want, is it?).

You are supposed to be impartial with respect to who produced which result. If you can't find a legitimate reason for your work to exist, other than your lack of preparation/prior knowledge, keep working.

With all that said, however, your approach does not need to be strictly better than known results to be of interest: maybe what you did is easier to run on a new dataset, maybe it would likely generalize better, maybe there is a reason to think it'd run faster. To answer any of that, you have to figure out how what they did works, how is it different compared to what you did and what other relevant research exists in the field. Normally, you learn all that before you do the bulk of your own research but it's normal to find out more as you get some elbow grease. Just do not bend the reality in order to not have to work on conclusions and future research.

3
  • Thanks, So can I mention both papers and say my approach is only better than one of them? or do I just need to address the one with higher accuracy? the paper with higher accuracy is for 2020 but the other one is for 2021
    – Narges Se
    Commented Oct 19, 2021 at 3:29
  • @NargesSe If I were you, I'd look for other research on the topic and compare to that as well, unless the topic is super narrow (sorry, don't know enough of NLP myself to figure it out on my own). So 1) What's done in NLP to do feature embedding/representation of sentences in general? 2) What's so special about rumors making it harder or making it possible to do a better job than when considering text in general? 3) What these guys did and what have you done and why do you think your approach may be beneficial to develop/employ?
    – Lodinn
    Commented Oct 19, 2021 at 3:40
  • Cite one or both but crucially, do find and describe other relevant work as well, even if it's less tightly connected to yours. Grabbing one or two articles and saying "we did what those guys did but better, also not really sure what's the place of our work in the grand scheme of things" is just no bueno.
    – Lodinn
    Commented Oct 19, 2021 at 3:43

Not the answer you're looking for? Browse other questions tagged .