Yiqin Fu Profile picture
12 Oct, 5 tweets, 3 min read
Spotify publishes the coolest research!

Facts from this paper: ojs.aaai.org/index.php/ICWS…

1) Everyone listens to music released in their adolescent years.

Also from nytimes.com/2018/02/10/opi… we know that the peak influence age for women is 13 and men 14.
2) Individuals across different states have the same level of diversity in music taste, but state-level diversity is quite different -- the measure doesn't correlate with the entropy of states' racial composition but does correlate strongly with the share of Hispanic population.
3) Jazz is a Pacific Northwest thing?
4) Moving to a different state doesn't change one's music taste. Things are set at age 13-14!

(The control group here is a matched sample of users who didn't move.)
So much interesting stuff here on recommendation systems, the diversity-engagement tradeoff, predictors of podcast popularity, the podcast-music listening tradeoff, female artist representation, etc.:


• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with Yiqin Fu

Yiqin Fu Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @yiqinfu

20 Sep 20
Trying to see if we can audit the Twitter cropping algorithm more rigorously... Here's the same person (Obama) with light and dark skin tones.

Photos taken from this paper: academic.oup.com/poq/article/80… ImageImage
Pair #2, same background, same pose

Photos from: uky.edu/AS/PoliSci/Pef… ImageImage
Pair #3, same background, same pose

source same as above ImageImage
Read 5 tweets
8 Sep 20
The Chinese magazine Renwu just published a long feature on “algorithms and food delivery drivers”.

Apparently the delivery platforms ask drivers to follow ***walking*** rather than driving directions, so drivers have no choice but to go against the traffic, run red lights, etc. Image
Drivers say it’s impossible to deliver “on time” if they follow traffic rules.

And as the algorithm gets trained on past delivery times, there’s a race to the bottom — the platform makes ever more unrealistic estimates, and drivers take on greater risk in order to stay in.
Rainy days are the worst because people place more orders.

Once, 30+ drivers in an area fulfilled 1000+ orders in less than three hours.

Drivers and regional hubs that fail to deliver on time or quit on rainy days get punished by the system.
Read 5 tweets
28 Aug 20
Exhibit A of social media as outrage merchants:

I share a lot of academic research on Chinese social media and frequently find that in the comment section, offensive one-liners get ranked higher than sincere discussions even though the former has fewer likes (see screenshot). Image
Weibo’s comment section has a Reddit-like nested structure, and the ranking algorithm takes into account **all** likes garnered by the parent comment and its children.

You can imagine how this kind of algorithm amplifies resentment and distorts our perception of public opinion.
I share a lot of gender-related research online.

Usually the top-ranked comment would be something offensive and unhelpful: “All women are...!”

This comment would then attract lots of counter-punches, making it even more attractive from the ranking algorithm’s perspective.
Read 5 tweets
27 Aug 20
Interesting discussions going around the Chinese Internet re the power WeChat wields as a super app that rolls your work, social, financial life into one.

Apparently when you get suspended, it becomes logistically difficult to pay back the loans you took out on the platform.
Toll payments in China are all made online these days.

One blogger said he missed several toll payments because his old WeChat account got suspended and the toll payment system wouldn’t accept a new WeChat account.

He worried that his financial creditworthiness may take a hit.
Per the discussions, once your account gets suspended, you can still log in, but you can’t send any messages.

In other words, you have to manually add your contacts when you create a new account.

There’s an appeals process, but I haven’t come across any success stories.
Read 6 tweets
26 Aug 20
A while ago I was playing around with some Chinese names data and found that people preferred shorter (one-character) names before the 1990s but that the trend was quickly reversed in more recent years.

I didn't think much of the finding, because it seems easily explained by..
Greater openness of society/marketization leading to more expressions of individualism.

But a friend pointed me towards a Chinese economist's blog post looking at name lengths in Chinese dynasties. Turns out that Ming and Qing saw similar trends!

(Chart credit goes to Chen Qin)
The source of the historical data is jinshilu, i.e. lists of top scorers of the imperial exam.

The economist Chen Qin also checked that the geographic distribution of the top scorers remained largely the same over time.

Here's the historical and the contemporary side-by-side:
Read 8 tweets
28 Jul 20
On Chinese social media people are sharing photos of their moms from the 1990s.

I’m no fashion designer, but I hope historians digitize these photos!! ImageImageImageImage
More: ImageImageImageImage
More: ImageImageImageImage
Read 6 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!