On Wed, 15 Apr 2015 17:34:18 +0530
Sarang Shrivastava wrote:

> https://www.google-melange.com/gsoc/proposal/review/student/google/gsoc2015/xlr_24/5629499534213120

I got "You are not logged in as the user in the URL".


> but I got some papers that prove that ANNs can be applied to spam
> filtering and they can be personalized depending from person to
> person.

I am a bit sceptical about neural networks because I've never
seen any evidence that's convincing. The following link that you
quoted is a good example of this: 

> http://www.ijert.org/view-pdf/3223/hidden-markov-models-and-artificial-neural-networks-for-spam-detection

Firstly the mail corpus looks to be a very easy one. It's from 2007 so
most of the spam is still pretty blatant, and the ham seems to be
entirely from technical mailing lists (without in-list spam). This
should be easy for any statistical filter, but the narrowness of the
ham makes it particularly well suited to a neural network.

Despite the corpus, the filter isn't doing all that well. The headline
figure of 93.91% comes with an FP rate of 13.73%. They don't quote
figures for any threshold with a good FP rate, but looking at the
graphs it doesn't look to beat the performance that Bayes gets on
much harder real-world mail. 

Reply via email to