Without getting into the specifics of this Project, I'd like to make a general observation, I just made on another post today.
Most of us didn't receive any formal education or training in communication or conflict resolution. Yet healthy communication is a skill most of us have to learn, just like reading and writing.
I'd like to suggest that we all consider investing proactively in ourselves, and our teams, to learn and practice healthy communication.
I believe such proactive learning can be a standard part of every team and organization, to reduce both the quantity and intensity of any conflicts. As well, such training provides effective tools for the team to deal with "conflict" in a constructive manner.
Huh... reading through their dataset of "toxic comments", there's an awful lot of subjectivity and borderline comments that paint an... interesting picture of what they consider to be toxicity. There are definite examples of hate speech in there, but a lot of them are fairly benign too. I have to wonder what an internet without those kinds of comments would look like...
Ouf - I was just reading the first bit, later on they get hilariously benign at some points. I'm not certain if there's some meme context I'm missing but "Feet are the bodies thermostat" and "Hope you have a great day!" seems pretty far from anything I'd consider toxic without a lot more context.
Could have fun constructing some toxic contexts for the "non-toxic" examples, too. Think of all the jokes that used the phrase, "That'll do, pig" which were so great because the original context was so lovely.
Github's CSV rendering is pretty bad when there are any wide columns. There's a column on the right that labels something "Toxic" or "Not Toxic" that is way off in right field for the web view.
This is obviously going to be subjective. I'd argue that people prone to using the currently popular buzz-word "toxic" when referring to opinions hold certain common political and social viewpoints and beliefs. In other words, this is implicitly a partisan (not objective) project that is mostly an attempt to enumerate wrongthink.
And antisocial behavior is “I don’t like this.” There are some cultures where loud chewing and burping is considered respectful.
I went to one such restaurant, felt personally disgusted because that’s the heigh of rudeness in my culture, and left the restaurant happy there’s a space where people can be comfortable with their native social norms.
Where going through an era of social upheaval, so I think the definition of anti-social is bound to evolve.
In different parts of the world I have experienced these two responses when getting in a Taxi: "Excuse me, please put your seat belt on or I'll have to ask you to get out" and "What about my driving offends you so much that you feel compelled to put on a seatbelt - I'll get you there safely."
Social constructs around rudeness and norms are quite variable and if you're going to be hopping around you need to set your mind to adapt[1].
Rudeness is defined by the intent - and the fact that burping (an action that can be almost entirely non-voluntary) is judged as such an offensive act is really a shame in western culture. If you give me a carbonated beverage I'm going to burp - I don't know why my digestion can't handle carbonated beverages at all but it's simply unavoidable so, while in western cultures, I'm pretty much unable to enjoy seltzer.
1. If you can - I know some folks with misophonia that have serious difficulties. Also, one of my coworkers a while back had a traumatic experience involving gum chewing.
And antisocial behavior is “I don’t like this.” There are some cultures where loud chewing and burping is considered respectful.
I went to one such restaurant, felt personally disgusted because that’s the heigh of rudeness in my culture, and left the restaurant happy there’s a space where people can be comfortable with their native social norms.
We’re going through an era of social upheaval, so I think the definition of anti-social is bound to evolve.
While I think the word toxic gets bandied around casually pretty often... having a read through their data set I think it's a pretty apt term for a lot of what I'm seeing. Most of these comments are utterly without value.
Though maybe a more appropriate name would be "The Flamebait Dataset" because, while I'm sure not all of these were intended as flamebait they could all quite quickly encourage some highly adversarial responses.
It seems to be largely focused around very recent American politics - well over 10% of entries contain the words "trump", "biden", "obama" or "brandon" (I just learned about this last one a couple of days ago, it's incredibly stupid and esoteric yet it makes up 3% of all entries).
And just eyeballing through a few entries it's very clear this is only targeted towards American toxicity - "fux news" and Alec Baldwin and Fauci and "LIBFUK DEMOCRAPS". This is not a useful dataset for any project that needs to filter beyond US-centric toxicity (to the extent that a dataset of 1,000 comments is useful at all).
Who on earth has appointed themselves as the arbiter of "toxicity"? Half the comments marked as "toxic" are just political opinions or sentences with bad language in them.
I tend to align with the person who suggested that the word "toxic" was just a secular isomorph of "sinful".
> Who on earth has appointed themselves as the arbiter of "toxicity"?
My goodness, relax. Nobody's writing a bill in the Senate or starting a new branch of the UN here. It's a CSV file with a few hundred crummy tweets. The selection is necessarily subjective and if you disagree with their choices, you can safely ignore them with zero consequences to you or anyone else.
The one thing that I can say definitely makes it a bad set is that it includes the popular phrase "Let's go, Brandon" by itself as a toxic statement.
For anybody still unaware, they were interviewing the winner of a NASCAR race on live television and the crowd was chanting "Fuck Joe Biden", the interviewer said that they were chanting "Let's go, Brandon". Because, you know, you can't acknowledge the swears.
This has become picked up as a sort of dog-whistle/shibboleth by those who still support President Trump. Even in places where saying "Fuck Joe Biden" is completely allowed. Which is any place that will let you swear.
Now, within the parameters of the data set, the statement "Fuck Joe Biden" would get classified as toxic. Just as "Fuck Donald Trump" would.
But, you do need the context to understand that "Let's go, Brandon" really means "Fuck Joe Biden" and is therefore toxic. If you train on this set, you could run the risk of confusing statements like "Let's go, Bradley" as toxic even though Bradley is just some dude. Or statements made to or about a guy named Brandon.
I usually don’t like those labels because they usually mean one of these things: 1) “I don’t like hearing this”, 2) “This might be true but it’s inconvenient as fuck” , or 3) “He is from the other tribe, let’s kick the living shit out of him!” Toxic, problematic. Those words. They mark “them” versus “us”.
But there’s a class of behavior you’d roughly classify as “if you see this, you kick the person from the party and never invite them again”. The designation “toxic” seems suitable. You feel filthy after even talking to some people and need to take a shower. I cannot define this precisely, but it’s the behavior that stops any productive discourse from happening, or puts off everyone in the room.
Toxicity isn’t defined, and it seems the dataset is mainly populated with inflammatory political statements.
I don’t really think you can judge toxic behavior accurately just by analyzing the words in individual comments. I have doubts that this dataset will be useful for “saving the Internet.” Honestly, I still don’t even think toxic comments are the main thing ruining the Internet right now.
I was really hoping this was about chemistry or biology...
Then again, 4 commits, 0 forks, 3 watchers... I don't really care what these people get their knickers in a twist about. Someone on the internet is offended, more news at 11.
Yeah, I was really disappointed. This is just a snapshot of someone's upvote/downvote history. The people upvoting this post should comment as to why this is remotely interesting, considering so far all the comments here are the same as ours.
Pretty interesting idea, as when you scan down the list, a theme that emerges is concrete declarative statements which lack originality and sophistcation, and include values/tribe-centric epithets, watchwords, and jargon.
I'd be concerned that such a corpus would automatically discriminate against people below a slightly advanced cognitive ability, and against people well above it who use irony. Then again, from the perspective of the makers of this, that may be a feature.
35 comments
[ 4.5 ms ] story [ 109 ms ] threadMost of us didn't receive any formal education or training in communication or conflict resolution. Yet healthy communication is a skill most of us have to learn, just like reading and writing.
I'd like to suggest that we all consider investing proactively in ourselves, and our teams, to learn and practice healthy communication.
I believe such proactive learning can be a standard part of every team and organization, to reduce both the quantity and intensity of any conflicts. As well, such training provides effective tools for the team to deal with "conflict" in a constructive manner.
https://hbr.org/2005/03/want-collaboration-accept-and-active...
Just a random sample of the top half of the set and I haven't run into any where I'm like "Eh, that's fairly benign".
I went to one such restaurant, felt personally disgusted because that’s the heigh of rudeness in my culture, and left the restaurant happy there’s a space where people can be comfortable with their native social norms.
Where going through an era of social upheaval, so I think the definition of anti-social is bound to evolve.
Social constructs around rudeness and norms are quite variable and if you're going to be hopping around you need to set your mind to adapt[1].
Rudeness is defined by the intent - and the fact that burping (an action that can be almost entirely non-voluntary) is judged as such an offensive act is really a shame in western culture. If you give me a carbonated beverage I'm going to burp - I don't know why my digestion can't handle carbonated beverages at all but it's simply unavoidable so, while in western cultures, I'm pretty much unable to enjoy seltzer.
1. If you can - I know some folks with misophonia that have serious difficulties. Also, one of my coworkers a while back had a traumatic experience involving gum chewing.
I went to one such restaurant, felt personally disgusted because that’s the heigh of rudeness in my culture, and left the restaurant happy there’s a space where people can be comfortable with their native social norms.
We’re going through an era of social upheaval, so I think the definition of anti-social is bound to evolve.
Though maybe a more appropriate name would be "The Flamebait Dataset" because, while I'm sure not all of these were intended as flamebait they could all quite quickly encourage some highly adversarial responses.
And just eyeballing through a few entries it's very clear this is only targeted towards American toxicity - "fux news" and Alec Baldwin and Fauci and "LIBFUK DEMOCRAPS". This is not a useful dataset for any project that needs to filter beyond US-centric toxicity (to the extent that a dataset of 1,000 comments is useful at all).
I tend to align with the person who suggested that the word "toxic" was just a secular isomorph of "sinful".
My goodness, relax. Nobody's writing a bill in the Senate or starting a new branch of the UN here. It's a CSV file with a few hundred crummy tweets. The selection is necessarily subjective and if you disagree with their choices, you can safely ignore them with zero consequences to you or anyone else.
For anybody still unaware, they were interviewing the winner of a NASCAR race on live television and the crowd was chanting "Fuck Joe Biden", the interviewer said that they were chanting "Let's go, Brandon". Because, you know, you can't acknowledge the swears.
This has become picked up as a sort of dog-whistle/shibboleth by those who still support President Trump. Even in places where saying "Fuck Joe Biden" is completely allowed. Which is any place that will let you swear.
Now, within the parameters of the data set, the statement "Fuck Joe Biden" would get classified as toxic. Just as "Fuck Donald Trump" would.
But, you do need the context to understand that "Let's go, Brandon" really means "Fuck Joe Biden" and is therefore toxic. If you train on this set, you could run the risk of confusing statements like "Let's go, Bradley" as toxic even though Bradley is just some dude. Or statements made to or about a guy named Brandon.
generalizing, i have heard it said by people who dislike Trump and think Biden's a disappointment.
But there’s a class of behavior you’d roughly classify as “if you see this, you kick the person from the party and never invite them again”. The designation “toxic” seems suitable. You feel filthy after even talking to some people and need to take a shower. I cannot define this precisely, but it’s the behavior that stops any productive discourse from happening, or puts off everyone in the room.
I think the word is fitting this behavior well.
I don’t really think you can judge toxic behavior accurately just by analyzing the words in individual comments. I have doubts that this dataset will be useful for “saving the Internet.” Honestly, I still don’t even think toxic comments are the main thing ruining the Internet right now.
Then again, 4 commits, 0 forks, 3 watchers... I don't really care what these people get their knickers in a twist about. Someone on the internet is offended, more news at 11.
I'd be concerned that such a corpus would automatically discriminate against people below a slightly advanced cognitive ability, and against people well above it who use irony. Then again, from the perspective of the makers of this, that may be a feature.
“500 comments”
I could scrape way more of that shit just from under one facebook post.
OTOH, I like how nonpartisan the list is, showing you can have bad apples no matter which color they are.
It is quite inclusive, including slurs against Democrats, Republicans, Covid conspiracy believers and even Christians.