5 ms·

TL;DR - If you post a lot of gray pictures of yourself with a sad face that nobody likes or comments on, you are probably depressed. They used the color of the

by jbob2000 9y ago

TL;DR - If you post a lot of gray pictures of yourself with a sad face that nobody likes or comments on, you are probably depressed. They used the color of the photographs, number of people in the photos, level of engagement, and happiness rated by Mechanical Turk to determine depression.

foxbarrington 9y ago
"The more comments Instagram posts received, the more likely they were posted by depressed participants"
- EGreg 9y ago
  In general I call this the Magarshak score: How many comments vs shares does a content post have? Comments - disagree with it, want to correct something, controversy Shares - agree with it, want to spread it more widely, echo chamber
  - jnbiche 9y ago
    Wouldn't that be the Magarshak ratio?
    - 5ilv3r 9y ago
      No because 100:100 and 0:0 would be equal in value but far different in meaning. It must be a score and not a ratio.
      - jnbiche 9y ago
        Yes, but that's not what this ratio purports to measure. Also, the fact that you wrote n:n with the colon in between would make it by definition a ratio.
  - foota 9y ago
    I think this works better for something like Facebook, Instagram comments tend to be compliments.
    - cgore 9y ago
      Yeah, it's really weird to me coming from the rest of the Internet to have random people I've never met say things like "Beautiful", "Great work", "I love it", etc., instead of, say, what you'll get in the comments of a youtube video.
      - imcrs 9y ago
        Sadly, if it's just a single compliment with no relation to the actual image posted, it's likely a bot doing the commenting. This is a common way for people to boost their follower count.
      - Retra 9y ago
        That's all I ever see on my employer's blogs and announcements, and is far more depressing to me than any honest criticism would be.
ge96 9y ago
Hahaha... mechanical Turk, I was one of those, the xxx ones were at least entertaining
- frandroid 9y ago
  What kind of tasks were you hired to do on this latter topic?
  - ge96 9y ago
    Tagging scenes Genders in the scene Positions etc... I'm curious how ML will do that, not how but accuracy would be cool to make (train) the model. Also with thumbnails that capture the viewer's attention. The other jobs were brute searching sites in hopes to improve SEO. Transcribing photos of receipts (funny those ads, some poor schmuck being paid 12 cents per 10 min job). Oh I remember an awkward one it was some social site's user photo uploads, I had to tag images with babies. So weird to see so many 'public' yet seemingly intimate photos. Not nude, just random people's photos. I did this for a whole week as much as I could do sanely, made $55 not worth it. Washing plates makes more money.
    - slig 9y ago
      > The other jobs were brute searching sites in hopes to improve SEO. I'm curious on how this works. Do you basically search for specific keywords and click on specific URLs in order to signal Google that the URL is more relevant?
      - ge96 9y ago
        That sememed to be the basis of it but this was a couple of years ago or more say 2010-2012 something like that man that was a while back wow. Yeah you had to search some term and find the link. It seemed to work as they would be beyond the first page result usually or at least not #1 but say #4 on page 1. But it could be biased you know based on your own search history I often check externally with say Tor to see how sites rank from other viewpoints than my own. Anyway hope that is helpful. You could also just sign up and look around. I think I still have a 99.99% satisfaction rate a lot of jobs you need something close to that to apply but it is shit work though. Some go up to $1.00+ woweee but those take like half an hour to do it's ridiculous.
jogjayr 9y ago
Wait so they didn't actually go back and see if the people they identified as depressed had a diagnosis of clinical depression?
- graphememes 9y ago
  What, you thought this was an actual scientific study?
  - jogjayr 9y ago
    Ok I went back and read more: "Data collection was crowdsourced using Amazon’s Mechanical Turk (MTurk) crowdwork platform. Separate surveys were created for depressed and healthy individuals. In the depressed survey, participants were invited to complete a survey that involved passing a series of inclusion criteria, responding to a standardized clinical depression survey, answering questions related to demographics and history of depression, and sharing social media history. We used the CES-D (Center for Epidemiologic Studies Depression Scale) questionnaire to screen participant depression levels [26]. CES-D assessment quality has been demonstrated as on-par with other depression inventories, including the Beck Depression Inventory and the Kellner Symptom Questionnaire [27, 28]. Healthy participants were screened to ensure no history of depression and active Instagram use. See Additional file 1 for actual survey text." So it sounds like they found depressed and healthy subjects via MTurk and verified them using surveys. I guess that's slightly more rigorous than just "people on MTurk thought the person in this photo looks depressed"
  - matthewmacleod 9y ago
    It literally is a peer-reviewed scientific study. Middlebrow cynical dismissals are boring.
    - nylonstrung 9y ago
      a peer-reviewed scientific study at the height of the replication crisis

foxbarrington 9y ago

"The more comments Instagram posts received, the more likely they were posted by depressed participants"

EGreg 9y ago
In general I call this the Magarshak score: How many comments vs shares does a content post have? Comments - disagree with it, want to correct something, controversy Shares - agree with it, want to spread it more widely, echo chamber
- jnbiche 9y ago
  Wouldn't that be the Magarshak ratio?
  - 5ilv3r 9y ago
    No because 100:100 and 0:0 would be equal in value but far different in meaning. It must be a score and not a ratio.
    - jnbiche 9y ago
      Yes, but that's not what this ratio purports to measure. Also, the fact that you wrote n:n with the colon in between would make it by definition a ratio.
- foota 9y ago
  I think this works better for something like Facebook, Instagram comments tend to be compliments.
  - cgore 9y ago
    Yeah, it's really weird to me coming from the rest of the Internet to have random people I've never met say things like "Beautiful", "Great work", "I love it", etc., instead of, say, what you'll get in the comments of a youtube video.
    - imcrs 9y ago
      Sadly, if it's just a single compliment with no relation to the actual image posted, it's likely a bot doing the commenting. This is a common way for people to boost their follower count.
    - Retra 9y ago
      That's all I ever see on my employer's blogs and announcements, and is far more depressing to me than any honest criticism would be.

ge96 9y ago

Hahaha... mechanical Turk, I was one of those, the xxx ones were at least entertaining

frandroid 9y ago
What kind of tasks were you hired to do on this latter topic?
- ge96 9y ago
  Tagging scenes Genders in the scene Positions etc... I'm curious how ML will do that, not how but accuracy would be cool to make (train) the model. Also with thumbnails that capture the viewer's attention. The other jobs were brute searching sites in hopes to improve SEO. Transcribing photos of receipts (funny those ads, some poor schmuck being paid 12 cents per 10 min job). Oh I remember an awkward one it was some social site's user photo uploads, I had to tag images with babies. So weird to see so many 'public' yet seemingly intimate photos. Not nude, just random people's photos. I did this for a whole week as much as I could do sanely, made $55 not worth it. Washing plates makes more money.
  - slig 9y ago
    > The other jobs were brute searching sites in hopes to improve SEO. I'm curious on how this works. Do you basically search for specific keywords and click on specific URLs in order to signal Google that the URL is more relevant?
    - ge96 9y ago
      That sememed to be the basis of it but this was a couple of years ago or more say 2010-2012 something like that man that was a while back wow. Yeah you had to search some term and find the link. It seemed to work as they would be beyond the first page result usually or at least not #1 but say #4 on page 1. But it could be biased you know based on your own search history I often check externally with say Tor to see how sites rank from other viewpoints than my own. Anyway hope that is helpful. You could also just sign up and look around. I think I still have a 99.99% satisfaction rate a lot of jobs you need something close to that to apply but it is shit work though. Some go up to $1.00+ woweee but those take like half an hour to do it's ridiculous.

jogjayr 9y ago

Wait so they didn't actually go back and see if the people they identified as depressed had a diagnosis of clinical depression?

graphememes 9y ago
What, you thought this was an actual scientific study?
- jogjayr 9y ago
  Ok I went back and read more: "Data collection was crowdsourced using Amazon’s Mechanical Turk (MTurk) crowdwork platform. Separate surveys were created for depressed and healthy individuals. In the depressed survey, participants were invited to complete a survey that involved passing a series of inclusion criteria, responding to a standardized clinical depression survey, answering questions related to demographics and history of depression, and sharing social media history. We used the CES-D (Center for Epidemiologic Studies Depression Scale) questionnaire to screen participant depression levels [26]. CES-D assessment quality has been demonstrated as on-par with other depression inventories, including the Beck Depression Inventory and the Kellner Symptom Questionnaire [27, 28]. Healthy participants were screened to ensure no history of depression and active Instagram use. See Additional file 1 for actual survey text." So it sounds like they found depressed and healthy subjects via MTurk and verified them using surveys. I guess that's slightly more rigorous than just "people on MTurk thought the person in this photo looks depressed"
- matthewmacleod 9y ago
  It literally is a peer-reviewed scientific study. Middlebrow cynical dismissals are boring.
  - nylonstrung 9y ago
    a peer-reviewed scientific study at the height of the replication crisis