@LorenzoC: Bayesian filtering actually wasn't my idea, I got it here: http://live.gnome.org/Epiphany_2fAdBlockExtension. I think they've made some good points there.
Now, if you would train the bayesian filter all by yourself, you would probably get lots of false positives, that's right. Idea is to release ABP with a predefined statistical database (Thundebird doesn't let you start from scratch either) and update the database regularly the way it is done with filter subscriptions. The user's corrections to the predefined database would be stored independently so the don't get overridden.
Ok, there are lots of ways how you could trick bayesian filtering. And once ABP can be used without even seeing its user interface it has good chances to become mainstream - meaning becoming a target as well. That was rue's argument against adding bayesian filtering last time I talked to him. But I think giving up without even trying is just stupid. And I don't see our chances all that bad if we manage to keep the statistics database up-to-date. Maybe we can achieve the latter by allowing users to send in their user's corrections database.
Adblock Plus 0.6 needs a new name
I like innovations so I vote for trying the bayesian filtering or any other idea that could lead to improvements, while I don't care at all about icons and such.Wladimir Palant wrote:@LorenzoC: Bayesian filtering actually wasn't my idea, I got it here: http://live.gnome.org/Epiphany_2fAdBlockExtension. I think they've made some good points there.
Now, if you would train the bayesian filter all by yourself, you would probably get lots of false positives, that's right. Idea is to release ABP with a predefined statistical database (Thundebird doesn't let you start from scratch either) and update the database regularly the way it is done with filter subscriptions. The user's corrections to the predefined database would be stored independently so the don't get overridden.
Ok, there are lots of ways how you could trick bayesian filtering. And once ABP can be used without even seeing its user interface it has good chances to become mainstream - meaning becoming a target as well. That was rue's argument against adding bayesian filtering last time I talked to him. But I think giving up without even trying is just stupid. And I don't see our chances all that bad if we manage to keep the statistics database up-to-date. Maybe we can achieve the latter by allowing users to send in their user's corrections database.
I don't know enough about it to understand well the pros and the cons, besides general and maybe meaningless reasoning.
I don't think that bayesian is more vulnerable than regex once the advertisers start to fight back, on the contrary it should be more versatile and adaptable. Of course it depends on how many elements you can "track down" to identify the ads from the rest.
I did not understand the subscription idea then (see the dev builds post).
The idea of getting the users inputs to tune the statistical database scares me a little also.
It was not me, just in case...LorenzoC wrote:I like innovations so I vote for trying the bayesian filtering or any other idea that could lead to improvements, while I don't care at all about icons and such.Wladimir Palant wrote:@LorenzoC: Bayesian filtering actually wasn't my idea, I got it here: http://live.gnome.org/Epiphany_2fAdBlockExtension. I think they've made some good points there.
Now, if you would train the bayesian filter all by yourself, you would probably get lots of false positives, that's right. Idea is to release ABP with a predefined statistical database (Thundebird doesn't let you start from scratch either) and update the database regularly the way it is done with filter subscriptions. The user's corrections to the predefined database would be stored independently so the don't get overridden.
Ok, there are lots of ways how you could trick bayesian filtering. And once ABP can be used without even seeing its user interface it has good chances to become mainstream - meaning becoming a target as well. That was rue's argument against adding bayesian filtering last time I talked to him. But I think giving up without even trying is just stupid. And I don't see our chances all that bad if we manage to keep the statistics database up-to-date. Maybe we can achieve the latter by allowing users to send in their user's corrections database.
I don't know enough about it to understand well the pros and the cons, besides general and maybe meaningless reasoning.
I don't think that bayesian is more vulnerable than regex once the advertisers start to fight back, on the contrary it should be more versatile and adaptable. Of course it depends on how many elements you can "track down" to identify the ads from the rest.
I did not understand the subscription idea then (see the dev builds post).
The idea of getting the users inputs to tune the statistical database scares me a little also.
You mean "Kindergarten". I'm sorry that I lost my temper, but it makes me mad if I hear this stupid questions and everything what has to do with Adblock Plus 0.5.11.2. Can't you stop them?Wladimir Palant wrote:I can recognize a troll when I see one. You are a troll, go back to your kindergarten.LorenzoC wrote:It was not me, just in case...