3 ms·

The article mentions a "groundbreaking study" from 2013 that a psychologist "spent years developing"... It sounds impressive, until I looked at the actual stud

by cs702 10y ago

The article mentions a "groundbreaking study" from 2013 that a psychologist "spent years developing"... It sounds impressive, until I looked at the actual study[0], which uses simple linear or logistic regression on 100 features obtained by low-rank approximation of an m users × n Facebook-likes matrix via plain-old singular value decomposition (SVD)[1], as shown here: http://www.pnas.org/content/110/15/5802/F1.large.jpg http://www.pnas.org/content/110/15/5802/F1.large.jpg I don't know how anyone could call this "groundbreaking," considering that lots of people in Silicon Valley have been using matrix factorization to predict consumer behavior and make recommendations for a couple of decades, give or take. I stopped reading at that point. [0] http://www.pnas.org/content/110/15/5802.figures-only http://www.pnas.org/content/110/15/5802.figures-only [1] http://www.ams.org/samplings/feature-column/fcarc-svd http://www.ams.org/samplings/feature-column/fcarc-svd

monk_e_boy 10y ago
I don't know much about US elections in regard to facebook and fake news, but I did pay attention during Brexit. Most 'news' was fed to us through newspapers (online and offline) the TV and radio. I didn't see a single fake news site, I saw some crap forwarded on facebook most of which had lots of comments calling it out as crap.
- dreamfactory2 10y ago
  there were a ton of brexit bots on twitter and in comment sections, they later started working for trump, and are now working for le pen. some stuff about it here - http://heatst.com/world/how-russias-twitter-bots-and-trolls-work-with-donald-trump-campaign-accounts/ http://heatst.com/world/how-russias-twitter-bots-and-trolls-...
- dredmorbius 10y ago
  Resurrecting an inexplicably dead comment: Ithere were a ton of brexit bots on twitter and in comment sections, they later started working for trump, and are now working for le pen. some stuff about it here - https://heatst.com/world/how-russias-twitter-bots-and-trolls-work-with-donald-trump-campaign-accounts/ https://heatst.com/world/how-russias-twitter-bots-and-trolls...
  - leereeves 10y ago
    Perhaps it's dead because it's dismissing people as bots and trolls because the author doesn't agree with them. > Broadly speaking, Trump has two categories of support on Twitter. Alt-right trolls, and Russian bot accounts
jgamman 10y ago
that seems a strange way to classify 'groundbreaking' - surely the result, even if preliminary and needing confirmation peer review replication etc, is what makes something potentially groundbreaking? typically, asking new questions is hard enough without requiring the latest buzzword acronym to be used in parallel.
privong 10y ago
> I don't know how anyone could call matrix factorization "groundbreaking." I stopped reading at that point. The article probably means the study was groundbreaking, not the mathematics. This is fairly common in science. The novel nature of the study may come from unique applications of existing technique, or applications of existing techniques to previously unavailable datasets. Focusing on the mathematical techniques used as an assessment of what is "groundbreaking" is a short-sighted way to approach it. For example, many (probably most) studies in astronomy/astrophysics aren't developing and applying new mathematical techniques as part of the studies. I think folks would agree that detection of exoplanets via radial velocity wobbles is groundbreaking. But that's "just" application of classical mechanics. The groundbreaking nature was the instruments with high enough precision to make those measurements and the teams who thought up the idea and made the observations happen. That being said, I'm not really qualified to assess the novelty of the study which is the subject of the article. But one should certainly look at more than just the tools used.
- cs702 10y ago
  People in Silicon Valley have been using matrix factorization for the same or similar tasks (like predicting behavior, making recommendations, and categorizing users) for ages.[0] I updated my comment to mention this. [0] For example, the Netflix $1 million competition in 2007 was won by a team that used an ensemble of methods, including a form of matrix factorization, to predict which movies people would like: http://www.netflixprize.com/assets/ProgressPrize2007_KorBell.pdf http://www.netflixprize.com/assets/ProgressPrize2007_KorBell... -- and using matrix factorization wasn't a new idea back then either.
  - optimuspaul 10y ago
    Again, not the methodology that is groundbreaking, but the findings, right? Even still why dismiss because you disagree with adjective used?
    - nostrademons 10y ago
      The big Silicon Valley tech companies (Google, Facebook, Amazon, etc.) have been using similar mechanisms for at least a decade to predict which ads people will click on and increase their likelihood of doing so. You can actually see what Facebook has predicted you're most interested in, based on their like/click/commenting/view behavior: https://www.facebook.com/ads/preferences/edit/ https://www.facebook.com/ads/preferences/edit/ This is the first reported case of these techniques being used to influence an election, but that doesn't mean it's the first time it was used to influence an election. It's very likely that Obama's campaign used similar techniques, considering it was staffed by Google/Facebook/Twitter data scientists who were certainly familiar with these techniques. It just appears in the press now because the Trump campaign & administration leaks like a sieve.
optimuspaul 10y ago
Explain for a five year old why you dismissed his study if you could. thanks!
- _rpd 10y ago
  It zooms past the now expected AI hype and doesn't stop until it is deep in science fiction. The correlation of Facebook-likes with personality traits is no more profound than the correlation of particular personality traits with watching NASCAR or buying opera tickets. Personalized political advertising/articles is at the same tech level as personalized advertising/articles in general. That is, still in its infancy.
caconym_ 10y ago
If the results are new and significant (in the qualitative sense) then it seems completely fair to call them "groundbreaking" even if they were arrived at without any sort of novel methodology. Whether they actually fit this description, I do not know, and I make no claim either way.
maus42 10y ago
It might be groundbreaking if no one else had the idea to use even plain old SVD to look at the problem. (And it would still look like a new method to some particular scientific subfield if the Silicon Valley ads and marketing people never published about the methods they are using in the journals these people read...) And learning SVD if you start at zero background in required maths, especially if nobody is exactly telling you what to learn and how to learn it and how to use it (yes I'm stereotyping the word 'psychologist' heavily here) could plausibly take "years". I'm doing maths and while I had first encounter with eigendecompositions during the 1st year linear algebra (which remarkably many students find hard), I think it wasn't until the 3rd year when we covered SVDs in their full theoretically detailed glory in the class. Lots of "groundbreaking" stuff involves 'simply' applying techniques developed elsewhere to new problem domains. There's still some low-hanging fruit left for applied maths and statistics. However, I don't know enough about this particular area to really say if the statement of "groundbreaking" are really warranted, but that could be one plausible defense of the claim. edit. Edited to apply to your edit.

by cs702 10y ago

monk_e_boy 10y ago

I don't know much about US elections in regard to facebook and fake news, but I did pay attention during Brexit. Most 'news' was fed to us through newspapers (online and offline) the TV and radio. I didn't see a single fake news site, I saw some crap forwarded on facebook most of which had lots of comments calling it out as crap.

dreamfactory2 10y ago
there were a ton of brexit bots on twitter and in comment sections, they later started working for trump, and are now working for le pen. some stuff about it here - http://heatst.com/world/how-russias-twitter-bots-and-trolls-work-with-donald-trump-campaign-accounts/ http://heatst.com/world/how-russias-twitter-bots-and-trolls-...
dredmorbius 10y ago
Resurrecting an inexplicably dead comment: Ithere were a ton of brexit bots on twitter and in comment sections, they later started working for trump, and are now working for le pen. some stuff about it here - https://heatst.com/world/how-russias-twitter-bots-and-trolls-work-with-donald-trump-campaign-accounts/ https://heatst.com/world/how-russias-twitter-bots-and-trolls...
- leereeves 10y ago
  Perhaps it's dead because it's dismissing people as bots and trolls because the author doesn't agree with them. > Broadly speaking, Trump has two categories of support on Twitter. Alt-right trolls, and Russian bot accounts

jgamman 10y ago

that seems a strange way to classify 'groundbreaking' - surely the result, even if preliminary and needing confirmation peer review replication etc, is what makes something potentially groundbreaking? typically, asking new questions is hard enough without requiring the latest buzzword acronym to be used in parallel.

privong 10y ago

> I don't know how anyone could call matrix factorization "groundbreaking." I stopped reading at that point. The article probably means the study was groundbreaking, not the mathematics. This is fairly common in science. The novel nature of the study may come from unique applications of existing technique, or applications of existing techniques to previously unavailable datasets. Focusing on the mathematical techniques used as an assessment of what is "groundbreaking" is a short-sighted way to approach it. For example, many (probably most) studies in astronomy/astrophysics aren't developing and applying new mathematical techniques as part of the studies. I think folks would agree that detection of exoplanets via radial velocity wobbles is groundbreaking. But that's "just" application of classical mechanics. The groundbreaking nature was the instruments with high enough precision to make those measurements and the teams who thought up the idea and made the observations happen. That being said, I'm not really qualified to assess the novelty of the study which is the subject of the article. But one should certainly look at more than just the tools used.

cs702 10y ago
People in Silicon Valley have been using matrix factorization for the same or similar tasks (like predicting behavior, making recommendations, and categorizing users) for ages.[0] I updated my comment to mention this. [0] For example, the Netflix $1 million competition in 2007 was won by a team that used an ensemble of methods, including a form of matrix factorization, to predict which movies people would like: http://www.netflixprize.com/assets/ProgressPrize2007_KorBell.pdf http://www.netflixprize.com/assets/ProgressPrize2007_KorBell... -- and using matrix factorization wasn't a new idea back then either.
- optimuspaul 10y ago
  Again, not the methodology that is groundbreaking, but the findings, right? Even still why dismiss because you disagree with adjective used?
  - nostrademons 10y ago
    The big Silicon Valley tech companies (Google, Facebook, Amazon, etc.) have been using similar mechanisms for at least a decade to predict which ads people will click on and increase their likelihood of doing so. You can actually see what Facebook has predicted you're most interested in, based on their like/click/commenting/view behavior: https://www.facebook.com/ads/preferences/edit/ https://www.facebook.com/ads/preferences/edit/ This is the first reported case of these techniques being used to influence an election, but that doesn't mean it's the first time it was used to influence an election. It's very likely that Obama's campaign used similar techniques, considering it was staffed by Google/Facebook/Twitter data scientists who were certainly familiar with these techniques. It just appears in the press now because the Trump campaign & administration leaks like a sieve.

optimuspaul 10y ago

Explain for a five year old why you dismissed his study if you could. thanks!

_rpd 10y ago
It zooms past the now expected AI hype and doesn't stop until it is deep in science fiction. The correlation of Facebook-likes with personality traits is no more profound than the correlation of particular personality traits with watching NASCAR or buying opera tickets. Personalized political advertising/articles is at the same tech level as personalized advertising/articles in general. That is, still in its infancy.

caconym_ 10y ago

If the results are new and significant (in the qualitative sense) then it seems completely fair to call them "groundbreaking" even if they were arrived at without any sort of novel methodology. Whether they actually fit this description, I do not know, and I make no claim either way.

maus42 10y ago

It might be groundbreaking if no one else had the idea to use even plain old SVD to look at the problem. (And it would still look like a new method to some particular scientific subfield if the Silicon Valley ads and marketing people never published about the methods they are using in the journals these people read...) And learning SVD if you start at zero background in required maths, especially if nobody is exactly telling you what to learn and how to learn it and how to use it (yes I'm stereotyping the word 'psychologist' heavily here) could plausibly take "years". I'm doing maths and while I had first encounter with eigendecompositions during the 1st year linear algebra (which remarkably many students find hard), I think it wasn't until the 3rd year when we covered SVDs in their full theoretically detailed glory in the class. Lots of "groundbreaking" stuff involves 'simply' applying techniques developed elsewhere to new problem domains. There's still some low-hanging fruit left for applied maths and statistics. However, I don't know enough about this particular area to really say if the statement of "groundbreaking" are really warranted, but that could be one plausible defense of the claim. edit. Edited to apply to your edit.