WoundedLamb: Voilà... false positive can only be reduced. Training can be done with the spam threads already created on NL so far (though the small size of the sample (data rows) can easily lead to overfitting if used continuously ). I'm only advocating for more features (data columns such as the logs mentioned above) just to bring up up dimensions a little bit without invoking the curse of dimensionality. I know how we feel when these apps take more info than we think is required but this one would be for a good course.
I don't think Nairaland take geolocation details, your browser would have detected that. Unlike major forums out there, Nairaland doesn't really streamline user experience to reflect their location. The argument would be that it doesn't need that seeing that it doesn't make location based suggestions. But then again, some of us out here sometimes get confused with the time display.
Like I said, Seun needs your help regarding these Chinese guys. Nairaland needs you, Madam Vicky. Please, help us! This forum belongs to all of us and not only Seun. Please!
Where you on Nairaland during the dark age (around 2014/15 when the site collapsed for days and everyone had to go to other social networks to ask questions)? Do you want that to happen again?
Correct! Scatter our head lol
Good point! In bold: exactly one reason I'd push back. Curse of dimensionality and law of trade-offs brouhaha. Sacrificing privacy for security has always been short-term resort. Short-term indeed
It does collect actually, to an extent though; not in the real sense of geolocation permission to pin down an exact location nor enough details to push through a model as this one. It'd require a mix of other things like the features you mentioned earlier.
Remember the application knows the IP addresses of it's users. So this can be used to narrow down your ISPs but not your exact location. Use of VPN also circumvents this effort though. So like I said, analyzing this independently, wont lead to making that much of a headway.
But I do think it's somewhat sufficient in addition to other data that'll be used. I mean it doesn't need a user's exact location to address this.
Oh and remember privacy law appears to be more of a mirage so don't be surprised if if if.......... let me rest my case.
The time location is a minute part of the series of confusion this site houses. And let's not even get into the nitty-gritty of how this forum manages users sessions. OMG Seun will soon ban my account. Lemme be going.
Na real dark age. lol. I was a member but it was one of those period I was off NL for a while so didn't realize until I saw threads about it. You can't guilt trip me with that.
So Oga/Madam, PLEASE dont deny this calling. I'm only here to cheer you on.
Please I want to use this opportunity to announce that our game show slated for today will be postponed until further notice. This is as a result of the none availability of my fellow coordinator (Maggilove). I just hope and pray that she's alright as her phone number has been switched off since Thursday and no response from her as regards all emails sent to her email address.
A new date for the game show will be announced later.
If anyone can help us reach Maggilove in anyway, please do as soon as possible. Thank you!!!
Bantino: Please has anyone made contact or heard from Maggilove?
Supervisor, thanks for the update.
I haven't heard from her and that's concerning as normally I'd still be able to reach her via emails but she hasn't responded to any of my mails. I hope and pray she's fine, and looking forward to have her back with us soon. Thanks.
WoundedLamb: Hhmm... great! Our leader is very much on top of this!
But are you suggesting training the system with live data? I think the Chinese would have finished building their empire before that yields any positive result! To really combat these guys using ML, Seun will need a pre-trained model with the weights stored in the DB. In other words, he needs to have a collection of already classified texts (spam and genuin) and pass them through the neural network till the optimal weights per node is calculated. He can then store this weights in the DB so that the system can use this to detect spam posts. Apart from the text itself, I think training with log details like users country (which will eventually be collected using HTML3 geolocation API on the live server), session time, etc. will help ensure reliability. But like I said, there'll still a be a very few genuine post that will fall through the cracks.
Yes, Seun might be working on this already but you can also reach out to him to make suggestions or even help out if possible
lol Build an empire kwa
Apt! Yes, I agree with you. Although there are average data to build the model, it will be time wasting to begin from scratch when there are existing ones to leverage. The pre-trained models is way to go and ofcourse there are several of them from other libraries, like earlier mentioned this isnt the first forum with such (except that these ones are lala's lol) so Seun wouldn't need to build one from scratch to clamp down on this.
I mean, the deltas will surely need to be mapped as it will never be a perfect model for this use case but better as a starting point and live data will need to be ingested to reduce the obsolescence.
Also note, pre-trained model would still not get rid of the false positives(genuine post errors). Having a model to build on is one thing but being able to train it to meet your needs is another.
Training with log details or session details isn't a bad idea either. I'd push back on that (for other reasons) if NL wasn't already collecting these info anyways. Moreover it's faster than building out a new model, but more susceptible/prone to errors.
Let me pretend I didn't see your last line. Team lead, this is your calling. Accept it wholeheartedly!
WoundedLamb: ML is a good suggestion and it will be relatively easy for Seun to implement cause there are already known networks that can handle such cases and Nairaland runs on Python (as I learned) so he'll only need to install the Anaconda framework to kick off. However, he will need to feed the ML network with more data than Nairaland presently collects if he really wants detect these guys. Again, it will still be somewhere tricky cause even with the best binary classification algorithms, genuine posts can still easily be classified as spam.
Agreed that genuine posts can still get classified as spam and those are the false positives that the forum will have to ruffle with for a while. But with detailed data ingestion and analysis, the predictive model will get better over time and the side effects can be minimal.
Also if it takes sacrificing some genuine posts for a while I think it's worth the sacrifice or worth the cause. (Of which those can be auto restored anyways) Learning, adapting, and improving on the pattern is really key to making it an iterative process.
I like to think the owner of this forum knows exactly what to do and maybe just maybe he's working on it but we're yet to see it's effect. I like to think so, else........!!!
Corbin: thanks for the clear up about you being here since 2015 it's nice, you too you are old
About the bolded I am not kidding let me refresh ur memory
Look at the old N and compare it to now and some other changes
If you go back to page 1 of this thread take note of how the quotes look like compared to now Mrs vickyrotex [cadet] please get ur facts right next time before you attack
Wahala dey o. What facts?
Wo joor I no fit laugh. Bwahahahaha!
When I disputed your statement, I meant something more substantial. You cant show me the same design with slight changes (so minute) and still refer to that as Alot has changed. Got it?
If saying you win makes you happy, you Mr. Cabin biscuit, WIN! and thanks for refreshing my memory