iiinova — where your digital self keeps evolving

Deen Freelon @dfreelon.bsky.social · 12/08/2026

If this kind of research is your jam and you're near the end of your PhD (on either side), there's still time to apply to our lab's open postdoc position www.asc.upenn.edu/news-events/...

asc.upenn.edu

Call for Applications: Postdoctoral Researcher in Social Identity, Politics, and Computational Methods

031

Deen Freelon @dfreelon.bsky.social · 12/08/2026

Well I try to make sure the content matches the headline!

Deen Freelon @dfreelon.bsky.social · 12/08/2026

It turns out that what TikTok displays for hashtags and keywords does not differ substantially between users in our tests, holding country constant

020

Well it sounds so simple when you put it like that 😆

100

I suspect the effects may be major in some cases and minor in others. We may very well conduct that study...

000

GitHub - dfreelon/pyktok: A simple module to collect video, text, and metadata from Tiktok.

I think the official API is limited to PhD students and up, and Apify costs money so... Pyktok! (The current version is in dire need of an update, which we're working on) github.com/dfreelon/pyk...

github.com

A simple module to collect video, text, and metadata from Tiktok. - dfreelon/pyktok

171

Of course with international users or one's individual TL, it could very well be a different story.

020

Interesting you should mention that--our paper doesn't get into this, but we've informally observed that for hashtags and keywords, there's little difference between what different users see. So the "new account view" ends up being pretty representative in our experience.

340

To clarify one thing--you might get some overlap between the API and scraping with very small hashtags. But our claim should hold for medium to large ones.

2132

This is a good guess but we have no way of verifying it, in part because we don't know how fast TT's moderation tech works. We're pretty sure scrapers offer a better lens on what users are more likely to see though.

150

Shoutout to @gayoung-jeon.bsky.social for spearheading this study--she did an amazing job. She's also lead dev on a near-complete rewrite of Pyktok, so watch for that in the coming weeks.

2192

As we note in the paper, scrapers pull from the user interface, which privileges popular videos. The official API says it's designed to pull chronological and random videos but our ability to verify whether it actually does this is limited

At any rate, how you get your TikTok data matters immensely. As always, pick the right tool for the job and make sure you understand its biases (as much as any non-employee can!)

1213

In particular, we found that the official API returned chronological and (maybe) random videos? (We have no way to independently verify claims of stochasticity... yay) While the scrapers tended to return much more popular samples (in terms of views/likes/comments).

1285

We didn't test this directly, but a study's findings could potentially shift from sig to non-sig to sig in the opposite direction depending on which data collection method the authors use.

3346

I wanna make sure everyone understands what this means: when you pull TikTok data for the same hashtag at the same time using the official API and a scraping tool, you'll get two completely different datasets.

38129

When we pulled data using the same sampling criteria and timeframe, we found zero overlap between the official API and the other two methods in most cases. No videos in common. 😶

25513

Deen Freelon @dfreelon.bsky.social · 16/07/2026

So our study compared three different data collection methods: the official TikTok API; Pyktok (a data collection tool of our own design); and Apify, a paid scraping service.

2276

Our lab's major influences include Klein & D'Ignazio's "Data Feminism", Bonilla-Silvia & Zuberi's "White Logic, White Methods", Tajfel & Turner's social identity approach, and Buolamwini & Raji et al.'s work on algorithmic discrimination. www.asc.upenn.edu/research/cen...

asc.upenn.edu

Politics, Identities, and Communication Lab (PICL)

We study how people’s various intersecting social identities (including race, ethnicity, gender, sexual identity, political ideology, religion, and national identity) influence how they communicate an...

091

Deen Freelon @dfreelon.bsky.social · 16/07/2026

I'm looking for folks interested in applying strong quant and computational skills to RQs involving race, gender, political ideology, religion, national identity, etc. If you consider yourself a "QuantCrit," you may be a fit.

151

Deen Freelon @dfreelon.bsky.social · 16/07/2026

My lab @ UPenn is hiring a postdoc to start in early 2027. I'm looking for someone with expertise in social identities, computational methods, and political communication: www.asc.upenn.edu/news-events/...

asc.upenn.edu

Call for applications: Postdoctoral researcher in social identity, politics, and computational methods

12228

Deen Freelon @dfreelon.bsky.social · 20/02/2026

Try it and LMK how it goes 🤭

012

Deen Freelon @dfreelon.bsky.social · 20/02/2026

This is solvable if publishers would do their jobs and manually verify every reference. Some do--most of my final copyedits on manuscripts tend to be about incomplete references (to genuine articles only, of course).

130

I mean you *could*, but should you?

000

I'd never thought about it this way before.

Yes, they show up on Google Scholar because they've been cited in published work (!)

131

In conclusion, if you're gonna use AI in your research, be very careful and please remain vigilant against this kind of epistemic contaminant. (Or, y'know, just write your articles the old-fashioned way.)

2280

Third, I'm disturbed by how this "technique" has been/could be used to attribute claims to scholars who never made them--possibly maliciously. A highly-cited incident could do a lot of damage if it's not discovered quickly.

1331

Second, what are these publishers doing?! Isn't decent copyediting part of their ostensible value-add? University libraries are shelling out an arm and a leg precisely to *not* have to deal with shit like this.

1354

First, it reflects incredibly poorly on the citing authors. You don't need to read every word of my articles (though I'd recommend you did), but please don't outsource your lit review to ChatGPT.

1392

It's probably as easy for AI to generate a whole fake article as it is to spit out just the reference, so...

100

These were almost certainly hallucinated by AI, and the whole thing is bad news for several reasons:

1261

Here are two of mine: scholar.google.com/scholar?hl=e... scholar.google.com/scholar?hl=e... . @chrisbail.bsky.social, did you know we recently coauthored a piece on the Russia-Ukraine war? Neither did I.

scholar.google.com

3285