You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Recently @pythonhacker and @harisibrahimkv reported few comments looks like a spam. Junction right now only allows logged in users to comment.
There are two levels of deducting/marking comment as a spam. During signup check the reputation of email/ip of user, if it is potential spam email id/ip, block the user, legit user can write back to the concerned person to activate the account.
One more way is to have a spam deduction based on the content. There are two ways to implement, one user can report this as spam, once threshold is crossed, differentiate it in UI and automatic spam detector can mark this as spam.
Once user has crossed the threshold of spam posting, account should be blocked.
Thresholding comments from same login ? If you posted a comment on a talk at time t , you cannot post another comment within t + delta_t where delta_t could be even an hour.
Not allowing postings from known spam IPs - there are well known lists we can scrub an internal SPAM IP list to.
Not allowing postings with less than n characters. This prevents postings like All the best Sir!, Wonderful Sir! etc. This forces people to think and write something. Possibly keep n at something like 120 or 140 chars.
How about just use a third party to evaluate the spammy nature of a particular comment. I suggest using askimet[1] if anyone doesn't have any objection with it.
I'm personally not in favour to including the spam detection logic coded into junction itself. If not askimet, then a publish pypi library that will give spam score for a comment should do.
Reading the docs suggests wordpress api key is required. I am sure, there will be other libraries/services, if some one can do quick prototype, it will be awesome.
First will collect comments across last 3-4 PyCons and using one of the libraries - get its spam detection rating - on each. Idea is to find out how much true and false positives are reported. Will update here with some libraries which are pure Python as much as possible.
On Thu, 6 Apr 2017 at 11:34 AM Anand B Pillai ***@***.***> wrote:
First will collect comments across last 3-4 PyCons and using one of the
libraries - get its spam detection rating - on each. Idea is to find out
how much true and false positives are reported. Will update here with some
libraries which are pure Python as much as possible.
I haven't seen any spam in last few years public comments. So it may be
better to rely some external dataset as well. But the application domain
here is proposal comments and but one of the available spam identification
dataset belongs to email category [0].
[0]: http://csmining.org/index.php/spam-email-datasets-.html
—
Activity
pythonhacker commentedon Jun 9, 2015
How about also,
kracekumar commentedon Jun 10, 2015
All these three should be implemented.
kracekumar commentedon Jun 10, 2015
The order for implementing above should
Point 3 -> Point 2 -> Point 1. We need to prevent spam account creation as well.theskumar commentedon Jun 15, 2015
How about just use a third party to evaluate the spammy nature of a particular comment. I suggest using
askimet[1] if anyone doesn't have any objection with it.I'm personally not in favour to including the spam detection logic coded into junction itself. If not askimet, then a publish pypi library that will give spam score for a comment should do.
[1]
https://github.com/miracle2k/python-akismet
https://akismet.com/
ChillarAnand commentedon Jun 15, 2015
akismet 👍
kracekumar commentedon Jul 5, 2015
Reading the docs suggests
wordpress api keyis required. I am sure, there will be other libraries/services, if some one can do quick prototype, it will be awesome.pythonhacker commentedon Apr 6, 2017
I am taking this up. Will try and get it ready for PyCon 2017.
pythonhacker commentedon Apr 6, 2017
First will collect comments across last 3-4 PyCons and using one of the libraries - get its spam detection rating - on each. Idea is to find out how much true and false positives are reported. Will update here with some libraries which are pure Python as much as possible.
[-]Spam deduction in Junction[/-][+]Spam detection in Junction[/+]kracekumar commentedon Apr 9, 2017
kracekumar commentedon Jun 22, 2017
Here is a csv file which has spam and non-spam content from junction.
ananyo2012 commentedon Mar 20, 2020
Unfortunately the spam data is gone from one of the recent changes. Need to dump it somewhere, maybe in a different branch