AI Researchers Made a Sarcasm Detection Model (venturebeat.com) 55
An anonymous reader shares a report: Researchers in China say they've created sarcasm detection AI that achieved state-of-the-art performance on a dataset drawn from Twitter. The AI uses multimodal learning that combines text and imagery since both are often needed to understand whether a person is being sarcastic. The researchers argue that sarcasm detection can assist with sentiment analysis and crowdsourced understanding of public attitudes about a particular subject. In a challenge initiated earlier this year, Facebook is using multimodal AI to recognize whether memes violate its terms of service. The researchers' AI focuses on differences between text and imagery and then combines those results to make predictions. It also compares hashtags to tweet text to help assess the sentiment a user is trying to convey.
"Particularly, the input tokens will give high attention values to the image regions contradicting them, as incongruity is a key character of sarcasm," the paper reads. "As the incongruity might only appear within the text (e.g., a sarcastic text associated with an unrelated image), it is necessary to consider the intra modality incongruity." On a dataset drawn from Twitter, the model achieved a 2.74% improvement on a sarcasm detection F1 score compared to HFM, a multimodal detection model introduced last year. The new model also achieved an 86% accuracy rate, compared to 83% for HFM. The paper was published jointly by the Chinese Academy of Sciences and the Institute of Information Engineering, both in Beijing, China. The paper was presented this week at the virtual Empirical Methods in Natural Language Processing (EMNLP) conference.
"Particularly, the input tokens will give high attention values to the image regions contradicting them, as incongruity is a key character of sarcasm," the paper reads. "As the incongruity might only appear within the text (e.g., a sarcastic text associated with an unrelated image), it is necessary to consider the intra modality incongruity." On a dataset drawn from Twitter, the model achieved a 2.74% improvement on a sarcasm detection F1 score compared to HFM, a multimodal detection model introduced last year. The new model also achieved an 86% accuracy rate, compared to 83% for HFM. The paper was published jointly by the Chinese Academy of Sciences and the Institute of Information Engineering, both in Beijing, China. The paper was presented this week at the virtual Empirical Methods in Natural Language Processing (EMNLP) conference.
Let me be the first to say... (Score:5, Funny)
Re: (Score:3)
Well, it's better than humans do...
Re:Let me be the first to say... (Score:4, Informative)
Whoosh....
OP was referencing this...
https://www.youtube.com/watch?v=EZ73Q4DwrGM [youtube.com]
Re: (Score:2)
Poor Frink, trying to do something for the world. Now he'll have to go the analog route [youtu.be] for the rest of that discussion.
Re: (Score:1)
Came for the Simpsons reference. Was not disappointed.
Re: (Score:2)
Re: (Score:2)
Indeed. Good luck with that.
Re: (Score:2)
Was that sarcastic, or just dumb?
Asking for a friend. Of course.
No? You don't believe I have a friend? Would you believe for a child in a big house that's painted all white?
"He whose name need not be mentioned."
Re: Let me be the first to say... (Score:1)
Re: (Score:2)
Re: (Score:3)
Boy I didn't see that one coming.
Re: (Score:2)
Oh, that's useful.
The sarcasm tag was accidentally left off of your post.
Re: (Score:3)
I never would have guessed someone would post that joke!
I'm sure (Score:2)
Re: (Score:2)
Was that funny? Or was that a sarcastic funny mod?
My humor detector seems to be in a recursive loop on this story. But many of the Chinese I've met seem to be lacking funny bones.
Re: (Score:1)
Can you detect my sarcasm? (Score:4, Funny)
Re: (Score:2)
I've been told there are times I say things where I'm being sarcastic, but people can't tell because of how I say it. It sounds like I'm being serious because I generally don't use any facial movements and my voice doesn't have that sarcastic tone.
Good luck on software trying to detect that sarcasm (he says non-sarcastically).
Re: Can you detect my sarcasm? (Score:2)
Have you considered external devices for display of emotions and vocal intentions? Perhaps a lapel pin that you trigger to blink "sarcastic" or "actually serious" as needed?
Re: (Score:2)
What A Relief! (Score:3)
No one posting on this thread needs to use "/s"!
Re: What A Relief! (Score:1)
Re: (Score:2)
I bet it does a great job! (Score:5, Interesting)
... but how does it handle Schrodinger's sarcasm, the kind where someone makes a statement that is halfway sarcastic and halfway nodding to a real issue?
Yeah, like good humour (Score:2)
twitter dataset? (Score:5, Funny)
"on a dataset drawn from Twitter."
sarcasm test(input) {
return True
}
Check your return type (Score:1)
I think you meant to return a boolean, not sarcasm.
Researchers in China (Score:2)
Oh what a surprise!
Re: Researchers in China (Score:1)
Well that's one layer of management done with (Score:2)
Apparently, I'm pretty sarcastic, and my two immediate bosses lacked this advanced AI detection capability.
It led to many a merry meeting for everyone else, and the younglings I was cross training in actual technology were apparently infected by my "attitude".
If it has to be explained.... (Score:2)
then it's no longer funny.
Re: If it has to be explained.... (Score:2)
If it was misunderstood, it was not funny in the first place. Tell a better joke.
Yeah, (Score:2)
right.
And so it begins! (Score:3)
Or as somebody once said, the danger of AI doesn't lie in AI outsmarting people, but it begins with AI outsmarting the dumbest of people.
Re: (Score:2)
Re: (Score:2)
Exactly! Sort of a kindler, gentler Cthulhu.
Re: (Score:2)
I believe it will be dumb people who are going to eat you and after AI has convinced them of your edibility.
Judge tries the accused (Score:2)
Training set (Score:2)
I'm glad to see that we are all contributing to its training set.
Should have built a bullshit-detector first (Score:2)
Just saying.
What a fantastic, useful accomplishment! (Score:1)
Xi is a wonderful, open, bright, and friendly person. I hope they clone billions of him so that the world will be a better place. He also resembles a friendly soft huggable yellow bear from popular youth literature. What more can you want out of a person; or bear for that matter.
I hate sarcasm (Score:2)
I hate sarcasm.[/s?]
If I said that both sarcastic and non-sarcastic would work.
Now what the difference between being sarcastic and being sardonic?
And why isn't sardonism a word?
Rejoice! (Score:2)
Chinese citizens under Commie rules shall now be able to sarcasm-check their contents before posting them online, so they can avoid arrests!
Oh wait, the checker has already called the police to turn you in.
is sarcasm an effable thing? (Score:2)
"sarcasm detection AI that achieved state-of-the-art performance"
So there must be a history of sarcasm detection and somewhere there is a rating of its progress. This new development has achieved the status of "state-of-the-art", which means it is equal to all the others at this point in time. Where are those others? Who is rating them?
For my part, I am often unaware of using sarcasm. People sometimes accuse me of it when I'm sure they're wrong. I used to do it a lot but I'm making a deliberate effort to st
Oooh goood. (Score:2)
Re: (Score:2)
Does it detect it on this: (Score:2)
"People's Republic of China"?
"I was being sarcastic" (Score:2)
"Well, I stole your face!"