🚨New Preprint 🚨
You might imagine that people with depression are less sensitive to rewards. But…maybe not on social media!
Across three Twitter datasets (over 17 million tweets from 7736 Twitter users), we find that depressed individuals show *higher* sensitivity to social
Do you want to use GPT for text analysis in R?
@steverathje2
and I filmed a 15-min tutorial on using the GPT API to perform text analysis tasks (e.g. sentiment analysis or emotion detection) in R.
So excited to announce that, starting this fall, I'll be pursuing a PhD at
@PsychPrinceton
with
@yael_niv
, delving deeper into computational psychiatry! Yay! 🥳🔎🧠
Had so much fun at
#cpconf2023
last weekend!
I presented early work suggesting that people with depression might be more sensitive to social media likes (with
@yael_niv
&
@ErikNook
).
Thanks to the organizers and all the attendees for super interesting discussions!
If you’re at
@SOBP
, come check out my poster about the relationship between how people group their experiences and psychiatric symptom dimensions. With
@yael_niv
,
@yeonsoon_shin
and the late
@sarahdubrow
. T153, today, between 5:30-7:30 pm!
Gave my first conference (data blitz) talk today! I can now confidently say that both anxiety and excitement scale with the number of attendees
#CPC2022
Presenting my work on latent-cause inference & psychiatric symptom dimensions today at
#RLDM2022
. Stop by poster 176 if you’re interested in computational psychiatry! W/
@yael_niv
@yeonsoon_shin
@sarahdubrow
Excited to present my work titled “Individuals with depression show heightened sensitivity to social media rewards” at my first
@SPSPnews
! I’m giving a data blitz as part of the Psych of Tech preconference + a poster in Session A today! Come say hi if you’re around
#SPSP2024
We have updated our pre-print on using GPT for text analysis.
Our most exciting new finding: GPT-4 Turbo (the updated GPT-4 model released this January) is even better than the prior version of GPT-4 at detecting psychological constructs in text:
Our paper on GPT for text analysis is finally out in
@PNASNews
!
We showed that GPT can detect different psychological constructs in different types of online text data across 12 languages. More recent versions were more accurate, particularly for lesser-spoken languages.
Many fields—including psychology, sociology, communications, political science, and computer science— use computational methods to analyze text data.
We find that the all versions of
#GPT
are capable of accurately detecting various psychological constructs in text across 12
New work co-led by
@steverathje2
and me: we find that GPT can detect psychological constructs in text across many languages, outperforms many popular text analysis methods, and may be a good multi-purpose tool for psychological text analysis
🚨 New Working Paper 🚨
Is GPT the future of psychological text analysis?
We test GPT’s (3.5 and 4) ability to detect psychological constructs in text across 12 languages, finding that it’s superior to many existing methods of automated text analysis
🚨 New Working Paper 🚨
Is GPT the future of psychological text analysis?
We test GPT’s (3.5 and 4) ability to detect psychological constructs in text across 12 languages, finding that it’s superior to many existing methods of automated text analysis
🚨We're hiring a lab manager!🚨 We're looking for a stellar post-bacc interested in studying emotions, supporting lab infrastructure, and building community. Please spread the word! See links in 1st comment for our mission and culture. Review starts 3/1!
..
Happy, proud and humbled to have graduated yesterday with a double First in both my MSci and BA! I am grateful to all the people who shouted at me “You can do it!” at times when I didn’t believe in myself anymore.
*you can sort of see my head in the second picture*
Cambridge: “shall I interest you in an MA degree you’ve done nothing for apart from being alive for 6 years since starting your BA?”
Dan: 😩🙏🏼🙏🏼🙏🏼 (especially as it means catching up with my UK besties)
Since depression is associated with lower reward sensitivity in the lab, you would expect the same on social media, right?
Quite the opposite! We consistently found that depression was associated with heightened reward sensitivity - participants with (more severe symptoms of)
There has been a lot of debate recently about how social media use affects mental health. Past studies have found effects that are small on average, and highly variable between individuals. The interplay between social media and mental health appears to be complex and
These quantified social media rewards might be preferentially sought by people with mental health vulnerability as a compensatory mechanism for less real-life social reward. At the same time, the variability in these rewards could be detrimental to mental health.
So excited to participate in the International Linguistics Olympiad 2019 as a volunteer! Throwback to high school when I would represent mother
#Romania
in this contest 🇷🇴🇷🇴
@IOLing_official
🇰🇷
Now In Press at Journal of Experimental Social Psychology!
With
@zakijam
and
@leorhackel
, we show across three field experiments (n = 1622) that seeing live theatre improves empathy, changes socio-political attitudes, and leads to prosocial behavior.
Cambridge University Press has just made all 700 textbooks currently available in HTML format on Cambridge Core free to access until the end of May to assist readers during the Covid-19 outbreak. This also includes 58 textbooks in Language and Linguistics.
Our results contrast with findings that depression is associated with blunted reward sensitivity.
This might boil down to the key difference in setting - people with depression might appear less sensitive to artificial rewards in lab tasks, but in reality may be more sensitive
We leveraged three large-scale datasets of Twitter posting data (over 17 million tweets total) linked with mental health information. The datasets varied in (1) size, (2) how the data were collected and (3) the depression measure used.
We defined “reward sensitivity” in this
The causal direction - whether depression leads to higher social media reward sensitivity, or social media reward sensitivity constitutes a risk factor for depression, or both - remains to be established in future work.
Enjoying this thread so:
Early primary: palaeontologist / ballerina
Late primary: egyptologist / astrophysicist
Early high: something sciencey, maybe dr/vet
Late high: ew no patients, just science
College: biology
Uni: developmental biology / genetics / let me stay in science
The mechanism we focused on was differential sensitivity of individuals to social rewards, such as likes, shares or followers.
Past work showed these rewards reinforce posting behavior, as predicted by reinforcement learning models.
At the same time, a rich body of literature
Thank you
@Sander_vdLinden
for being such an incredible supervisor!!
And Jamie Druckman &
@leedewit
for all the thoughtful questions in the PhD Viva.
& Congrats
@RakoenMaertens
!!
I am Dr. Steve now 🤓
@steverathje2
@RouxCaroline
@jayvanbavel
Ultimately, using GPT for private data should be included in IRB protocols and judged by the IRB for risks vs. benefits like other study components.
In the paper, we show that there are large individual and dataset differences, which might be explained by differences in the overall receipt of likes.
Nonetheless, the effect of depression is *robust* across datasets (despite the different populations and depression measures
This work also establishes a mechanistic link between depression and a specific feature of social media platforms - the social rewards, which deviate from real life rewards in that they are (1) quantifiable and (2) more variable due to algorithms that create “virality”.
@theotheredmund
@steverathje2
We also tried giving GPT examples to improve its performance (known as ‘few-shot learning’). You can find more details in the paper, but we essentially found the results were variable and performance didn’t always increase over the ‘zero-shot’ case (no examples)
How did language on Twitter change from February to March?
I created a comparison word cloud to examine the popular words on United States Twitter in February (Blue) vs. March (Red).
As you can see, things changed very fast...
@siminadragos26
or when you think the supposedly wonder underdog candidate (who is actually p shady) will obviously get into the second round but doesn't
#babeletotcupsdvoteaza
@theotheredmund
@steverathje2
Regarding temperature, we fixed it to 0 to get GPT’s most likely output. It would be really interesting though to run it with higher temperature (more exploration) and look at the variance in the responses, and compare that with human variance!
@theotheredmund
@steverathje2
Thank you!! Yes, we experimented a bit with prompts on a subset of the sentiment and discrete emotions data. One thing we found is that GPT works much better if you ask it to provide numbers rather than full labels for a classification task.
Do you want to use GPT for text analysis in R?
@steverathje2
and I filmed a 15-min tutorial on using the GPT API to perform text analysis tasks (e.g. sentiment analysis or emotion detection) in R.
Same-Sex Sexual Attraction Does Not Spread in Adolescent Social Networks (2015). Although both sexual activity (OR=1.79) & desire to have a romantic relationship (OR=2.69) may spread from person to person, attraction to same sex partners did not (OR=0.96)
@AlexandruZanca
but it’s a pretty fundamental finding from the 80s - I’m pretty sure we learned about this in high school - Vox just tries to reproduce basic human experiments to help explain their results to people, which I think is a nice initiative
@simongerman600
Someone conflated Austroasiatic (Vietnamese, Khmer) with Austronesian (Indonesian, Tagalog etc.). Completely different language families.
Thoughtful & well-written article by
@steverathje2
on how people react to different kinds of 'common enemies' and how internet might change our natural response to this one
Everyone: the UK can't possibly make a worse decision than Brexit
Corona: *hold my beer*
At least they can interfere with each other so we might see some good coming out of this
Researchers speak out in the largest ever survey of
#ResearchCulture
. It's time for change, and everyone can help
#ReimagineResearch
👉 See the results and help decide what happens next:
@siminadragos26
yasss! the teenage love triangle songs make me cry like I used to listening to Red/Speak Now 😢😭 i was a bit skeptical at first but it really is a beautiful album beginning to end
🚨New Preprint 🚨
You might imagine that people with depression are less sensitive to rewards. But…maybe not on social media!
Across three Twitter datasets (over 17 million tweets from 7736 Twitter users), we find that depressed individuals show *higher* sensitivity to social