[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: (off-topic) self-plagiarism
On Fri, 10 Jul 2009, Joe Sollini wrote:
> Sorry to bring this up again but having had a look for through these papers
> Instead of finding six papers I was only able to find 3 (but six links
> to papers).
I found seven papers with virtually the same abstract (the one I received
for review is the 8th). Unfortunately, I have access to the full text only
in 3 cases (plus the one for review, but I keep that one in secret...), so
this is why I have to judge based mainly on the abstracts.
(I can send you a list, maybe you can help me get the remaining ones.)
> If he has a model with a wide range of applications and
> applies this model to fields as disparate as face recognition and virology
> it could perhaps be deemed as fitting that they need to be published in
> journals that people practising in the respective fields read?
I definitely agree with that. But this would require the content of the
paper to be:
1. We claim that we have a new theoretical model and that it is
applicable to the field of the journal (conference).
2. The description of the newly proposed model.
3. Empirical justification on data taken from the specific field.
This doesn't hold in this case, as I'll explain below.
>From the 7 paper titles of 3 says "a fast ... model", and 4 ones go like
"a fast model applied to the field of ...". So, the titles themselves accord
with your description above. However, let's move on to the abstracts.
>From the 7 abstracts 5 starts with the sentence: "This paper
presents a new approach to speed up the operation of <model>".
So the topic of the papers (according to the abstract) is NOT the
application of the model to a new domain, but a theoretical result on how
to compute it faster than earlier. Although the other two abstracts start
with "this paper presents an intelligent approach to detect...",
the remaining text is the same in all seven cases: "it is proved
theoretically and practically that the number of computation required
<by the new method> is less than that needed by the <old method>. So
although the paper titles claim the the method will be tested on a new
domain, there is no word about that in the abstract!
Theoretical chapter of the three papers: these are word-by-word the same
in the 3 papers I have access to. Notice again that the formulas are about
the speed-up factor (number of operations required) of the method compared
to the old one, so these again agree with the abstract, not the titles.
Now, the funniest part: the experiments. In two of the papers the numbers
are given in diagrams, in one in tables, so I cannot really tell if they
are different or the same. However, these results are clearly ALL about
speed-up ratios. So while the titles say that we will apply the method to
a new domain (virus detection, code detection, record detection, etc.).
there are NO detection results given at all! Just speed-up results,
(as promised by the abstract). No proof that it works, only proof that it
can be faster than before. And the most shocking part: none of the papers
says ANYTHING about the test data! Only that these are Matlab simulations.
But it is not stated at all that the data were domain-specific. I simply
can't believe that these went through a review process. Ah, and finally, the
Conclusions: its again the same in all papers, stating that "computations
have shown that <new model> requires fewer compuation stepts that <old
model>". Which is true, but thas nothing to do with the claim of the
titles that the model will be applied to a new domain.
Again, I can't say anything about the remaining 4 papers, but based on
their abstracts I suspect that they were also "generated" with the "let's
adjust the title and the first sentence to the domain" method. (which,
as I said, would be acceptable if the experiments were also adjusted...)
Hungarian Academy of Sciences *
Research Group on Artificial Intelligence * "Failure only begins
e-mail: tothl@xxxxxxxxxxxxxxx * when you stop trying"