Kevin Yang
@yang3kc
Followers
1,582
Following
1,161
Media
520
Statuses
2,952
Kaicheng Yang, PhD | Postdoc @Northeastern | Computational social science, AI, bots, misinformation, + | views mine
Boston, MA
Joined May 2019
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Haitian
• 593459 Tweets
Springfield
• 427110 Tweets
Labour
• 373783 Tweets
Yunes
• 242065 Tweets
Argentina
• 225648 Tweets
Colombia
• 203286 Tweets
Morena
• 202474 Tweets
PS5 Pro
• 188817 Tweets
Senado
• 181133 Tweets
PlayStation
• 144481 Tweets
Anna
• 134947 Tweets
Sony
• 134683 Tweets
England
• 117094 Tweets
CONGRATULATIONS BTS
• 88669 Tweets
Nintendo
• 74282 Tweets
Tony
• 73586 Tweets
#الكويت_العراق
• 64390 Tweets
バーレーン
• 51105 Tweets
Swalwell
• 50858 Tweets
Marko Cortés
• 26676 Tweets
#TemptationIsland
• 26417 Tweets
Finland
• 23330 Tweets
Trent
• 22878 Tweets
Lola
• 22589 Tweets
Barranquilla
• 20277 Tweets
Fang Fang
• 16938 Tweets
اليابان
• 16887 Tweets
Dave Grohl
• 16393 Tweets
Bernstein
• 15572 Tweets
Harry Kane
• 15265 Tweets
Adult Content
• 15223 Tweets
#KohLanta
• 14392 Tweets
Alfred
• 11673 Tweets
De Ligt
• 10946 Tweets
#NEDGER
• 10261 Tweets
Last Seen Profiles
This is not good, "Surprisingly, we observe a significant decline in LLMs’ reasoning abilities under format restrictions."
Link:
40
104
574
In other news, I married my best friend and the love of my life today. We have been dating for 7 years, most of which have been long-distance. Every reunion, including this one, was not easy. But we made it. And I'm the happiest person on earth.
44
2
272
Found this powerful scraping tool lately. It's a browser extension, so it works on all web pages you can visit and no coding is needed. Below is what it looks like when I try to scrape my Twitter timeline. It works especially well on tables. 1/2
5
36
160
🚨New dataset🚨
Studying online discussion about the 2022 US midterm elections? Check out our dataset MEIU22, a collection of social media posts from multiple platforms.
Paper:
Data+code:
1/5
3
54
155
Some personal updates: I just joined
@davidlazer
lab as a postdoc at
@Northeastern
. Exhausted due to moving & settling in, but quite excited about this new chapter.
If you are also in the Boston area, feel free to reach out, we can grab a coffee together!
17
4
150
TikTok announced that researchers can access more account-level data through its API. Although the decision-making process isn't transparent, it seems they are taking suggestions and requests from researchers, including us.
1/2
1
35
134
Fake accounts like those below use profile pictures generated by AI (GANs)! Our new study reveals their characteristics and estimates their prevalence on Twitter: about 10k daily active users!
With
@danishjeetsingh
and
@OSoMe_IU
1/7
7
48
111
Excited to share that our work "Characteristics and prevalence of fake social media profiles with AI-generated faces" has been accepted by
@journalsafetech
. Moreover,
@danishjeetsingh
will present our poster
@IC2S2
on Jul 19, be sure to check it out!
1/3
1
25
91
Some personal news: I passed my PhD defense yesterday. Still need to finalize my dissertation and some paperwork, but I guess you can call me Dr. now 😄😄😄.
#PhDone
Many thanks to my amazing advisors:
@OSoMe_IU
@yy
@RealBreaPerry
Alessandro Flammini (not on Twitter).
1/2
2
6
33
15
7
88
I've been using ChatGPT to help me filter new arXiv papers every day for two months. It saved me a lot of time and allowed me to stay on top of the field's development. In case you want to try:
Here is how it works: 1/5
1
18
67
Heading home from
@IC2S2
. One of the best conference experiences 😊. Met many old friends and made new ones. Very encouraged to see participants from more diverse backgrounds and younger generations joining the CSS community.
(Quite lucky that my flight was not affected)
3
1
66
Check out our paper "The COVID-19 Infodemic: Twitter versus Facebook", just published in
@BigDataSoc
Link: (open access)
With an amazing team:
@frapi9
@patrick_pmhui
David Axelrod
@ChrisTorresLugo
@jb_tweets
and
@OSoMe_IU
0
27
62
Just appeared on Yahoo finance live sharing our views on bot detection using
@botometer
and the Twitter/Musk lawsuit
0
10
60
Those doing computational social science should read this.
Three Gaps in Computational Text Analysis Methods for Social Sciences: A Research Agenda
Link:
2
19
57
Got reminded that I wrote this tool a couple years ago as I went through my CV. Still think it's cool and useful, so here is the link:
1
4
54
Some late-night talking, things I usually wouldn't tell people. Although I consider myself doing ok in my career as a researcher, I still feel anxious almost all the time. I have panic attacks from time to time and trouble sleeping most nights. 1/3
4
0
50
After seeing more and more "verified" bot accounts on this platform, many wonder who would have the money to pay for the premium fee. Turns out they don't need to! If bots can create enough content impressions, they get paid! Pic below shows an operator bragging about this.
1/4
3
7
49
Twitter API updates:
1. Free tier -> $40k/month (yep, there is a k)
2. $100/month tier has very limited functionalities
3. Happends in 4-6 weeks
4. Decahose will prob be the only option for data collection
5. Based on conversation with Twitter enterprise rep
Source 👇
Just got news that the new Twitter API has an ENTRY price of $40k/month.
It'll probably kill the long tail of tools/services that were /are crucial for Twitter's network effect success.
17
68
207
5
22
46
Excited to share that our paper "Anatomy of an AI-powered malicious social botnet" with
@OSoMe_IU
has been published in
@journalqd
.
DOI:
I'll present it
@ICWSM
on June 4 at 10am and the poster session in that afternoon.
See the thread below for details:
0
9
44
1
3
40
The prevalence of social bots has been at the center of the dispute between Twitter and Elon Musk amid Musk's takeover. But how does the public perceive the issue of social bots? Our new working paper sheds some light.
🧵 1/7
Link:
2
18
37
Hot take, all CS, IS background computational social scientists should take some serious social science classes
3
1
39
New Human Communication Research paper: Algorithmic Agents in the Hybrid Media System: Social Bots, Selective Amplification, and Partisan News about COVID-19
Link:
With:
@ZeningDuan_Ze
@TheJianingLi
@JosephineLukito
Fan Chen
@dvshah
@sijiayang_camer
2
10
36
FBI took down an AI-powered bot farm recently. Two examples provided are clearly using GAN-generated faces, similar to those documented in our recent paper . The 93-page affidavit describing the identification process is worth reading. 1/2
1
12
34
Great to see this work finally out congrats to
@btrantruong
@OSoMe_IU
!
A social media model suggests that bad actors can make low-quality content go viral most efficiently by infiltrating a community and gaining followers rather than by just “flooding the zone” with junk. In PNAS Nexus:
1
31
109
2
8
35
In light of Meta's decision to discontinue CrowdTangle, we,
@transparenttech
and
@GWIDDP
, are conducting the DSA Data Access Audit. If you work with social media data in some way, please consider filling out the survey. Details:
1
18
33
Excited to announce our paper "Can GPT-4 Models Detect Misleading Visualizations?" to be presented at
@ieeevis
.
Link:
With
@ali_sarvghad
and talented undergrads Jason Alexander and Priyal Nanda.
1/4
1
4
35
We at
@transparenttech
wrote an open letter to call on Twitter to keep their API free to access.
Read the letter at
1/2
3
20
34
Need to process large amounts of data with ChatGPT but struggle with the API calls? Check out my tutorial at
Currently covers the topics below w/ instructions and example code. If you find it helpful, plz share and star. Issues and PRs are welcome!
0
8
33
Checkout our new New media & Society paper “Asymmetrical perceptions of partisan political bots”
With
@YyjHarry
@OSoMe_IU
@JamesShanahan
DOI:
Free preprint:
1/3
2
13
31
I'm very thrilled to receive the 2021 Luddy Outstanding Research Award from
@IULuddy
. I want to thank the school for creating such a wonderful environment that fosters innovation and collaboration and to thank my mentors and collaborators. They made whatever I'm doing possible.
3
2
31
More media coverage on how the change of Twitter API could affect the research community (I'm quoted)
2
17
31
Not to insinuate causality, but it appears the botnet was suspended by Twitter a couple of days after the release of our paper 😅. Before that, the botnet had been active for over 6 months undetected.
0
7
31
Can finally reflect on the post-API conference. Very nice gathering and lots of stimulating discussions. My takeaways:
1. All Twitter researchers are (unequally) screwed
2. Some emerging alternative solutions, but no silver bullet
3. Some good changes on the horizon
1/2
2
1
30
I don't want to be too mean, but some computational social science research is neither computationally sound nor socially interesting.
1
1
28
2023 was a special year. Finally got PhD and visited home for the first time after years of grinding. Finally got COVID and boy it hit me hard. Then I moved to Boston to start a new journey. Had the lowest point of my life, but also many highlights.
#HappyNewYear
2
0
27
Had a great time at
@UMassAmherst
today sharing my takes on bots and bot detection. (I think Ethan did a better summary of my research than myself 😆)
Great talk at
@UMassAmherst
today: Kai-Cheng Yang of
@Botometer
on "Fantastic Bots and Where to Find Them",
0
12
38
1
2
27
The team has been analyzing data we collected from Twitter, Meta, and Reddit in 2022. Just realized that the same kind of data collection and analysis will no longer be possible 😢
5
0
26
It's confirmed, the academic research API access for Twitter will be gone after Feb 13 🪦
It will be replaced by the $100/month "basic access." Details coming on Monday.
#NoResearchWithoutAPI
2
11
26
Most LLMs cannot guarantee the output is valid JSON, even in JSON mode. A python package called "json_repair" could be really handy in these cases:
Wrote a unified interface for querying LLMs from different providers. Currently supports OpenAI, Google, and Together (various open-weight models).
Link:
0
0
15
2
0
26
Trying to apply for TikTok's academic API access, they actually ask me to fill out a Word doc and email it back to them 🤯🤯🤯
3
0
25
BTW, I'll be serving as the publicity chair for ICWSM24. Please follow
@icwsm
for the updates, and feel free to DM if you have any questions regarding the conference!
The ICWSM24 organizational committee is working hard to prepare for the next conference at Buffalo. In the coming days, we will announce the logo, website, a new global south initiative, CPF, and the exact dates for the conference. Please stay tuned!
0
12
42
0
2
24
Happening now
@IC2S2
"Models for data access in the post API world" by
@davidlazer
@libbyh
@TaliaStroud
@RebekahKTromble
0
3
25
Friday at
@IC2S2
, my lovely coauthors will present 2 posters and 1 talk:
- 1:30PM: Effect of LLM-generated fact-checking
@mdeverna2
- 1:30PM: New metric for localness of news domains
@Pranav__Goel
- 2:30PM: Problems with social media polls
@przemyslslaw
1/2
1
4
25
Asked by some journalists if it's possible to estimate the prevalence of bots on Twitter the other day. I actually had some not perfect but reasonable ideas that all start with scanning millions of tweets/accounts.
Now you tell me Twitter did it by looking at 100 accounts?
2
4
24
Forgot to mention that these bots also leverage ChatGPT to produce content and replies. The paycheck from Twitter could easily cover the cost of API calls to OpenAI.
After seeing enough instances, I believe a part of this platform has already turned into a Westworld.
4/4
1
2
23
Checkout our new preprint 'Neutral Bots Reveal Political Bias on Social Media'
Collaboration with Wen Chen,
@diogofpacheco
@OSoMe_IU
.
Link:
1/8
2
12
23
Have a Twitter project that was canceled or changed? We'd love to hear from you!
@transparenttech
is running a survey to understand how research is impacted by changes to Twitter. Here is the link:
#research
#socialmedia
#twitter
2
17
20
Friday at
@IC2S2
, we will present two papers on AI-powered bots/fake accounts on social media:
- Plenary talk at 9AM: Anatomy of an AI-powered malicious social botnet
- Poster at 1:30PM: Characteristics and prevalence of fake social media profiles with AI-generated faces
1/2
3
6
22
After listening to some friends' stories, I realized how lucky I was during my PhD. The journey was not easy, but there wasn't a single moment that I didn't feel supported.
0
0
20
My guess, this "basic access" is not enough for most research projects, the $100 monthly fee is too high for 95% of the researchers.
If we want to keep the whole Twitter research body and community alive, we will have to start considering alternatives.
#NoResearchWithoutAPI
1
3
22
Another nice thing is that you could upload a list of URLs and have the tool scrape the same elements from them automatically. Can't believe I was writing scrapers myself not long ago 😅
Link to the extension:
Works for all Chrome-based browsers.
2/2
2
2
21
The findings empirically show that AI-generated faces have been deployed at scale to create fake accounts. Earlier, we showed that ChatGPT has been used to enhance bots. Apparently, AI-powered fake accounts are here to stay. Are you ready for it?
6/7
2
5
20
1
4
21
Shared my thoughts on how AI might meddle with the election year with
@TIME
. Check out the video:
1
5
20
Twitter might lack the incentive to deal with these accounts because it's really a win-win situation: Bot operators make money while Twitter makes money and boosts DAU. Why bother?
But who loses?
3/4
2
1
20
@jsrailton
@chrismohney
I wrote about their business model lately
After seeing more and more "verified" bot accounts on this platform, many wonder who would have the money to pay for the premium fee. Turns out they don't need to! If bots can create enough content impressions, they get paid! Pic below shows an operator bragging about this.
1/4
3
7
49
1
4
18
Happy to see that Musk's lawsuit against
@CCDHate
was dismissed by the court and that the ruling cited our survey with
@JosephineLukito
@m_dot_brown
. But reading the ruling gave me a lot to think about
1/3
1
6
20
@ChrisTorresLugo
and I will demo how to use OSoMe tools () to identify social media manipulation at 2022 Computation + Journalism Conference (). We have some cool news tools to show! You can join remotely and registration is free!
0
10
20
One lesson I learned the hard way: don't try to write your CV (or personal website) right before the job application. Create one RIGHT NOW and iterate it regularly. Add a new line when you have a new achievement to remind yourself you are progressing. 1/2
1
1
18
Something I have been doing increasingly is writing the manuscript, preparing the slides, and running the data analyses simultaneously. It helps me think about the project end-to-end and reminds me of the big picture while I'm dealing with details.
1
0
17
Keynote by
@rzafarani
: Fake news detection: Limited Ground Truth, Limited Text, No Understanding of Spreading Intent
#CySoc
Also on Zoom, DM me for link!
1
3
17
Checkout our new preprint "Prevalence of Low-Credibility Information on Twitter During the COVID-19 Outbreak". Collaboration with the amazing
@ChrisTorresLugo
,
@OSoMe_IU
Link:
1/5
1
12
17
🙌 It's a massive collaboration that took an OSoMe (pronounced awesome) village:
@mdeverna2
@RachithAiyappa
Wanying Zhao
@manitapote
@btrantruong
and many others
5/5
0
6
17
2) Conservative personas are more confusing and Republican participants perform less well
3) Republican users are more likely to confuse conservative bots with humans, whereas Democratic users are more likely to confuse conservative human users with bots
3/3
1
6
16
Participated in a discussion that was live broadcasted. While I was talking, my mom sent me a message: "lower your head, your chin looks enormous."
2
0
16
Botometer is back! Well... sort of...
Introducing Botometer X! The original Botometer website was disabled after Twitter (now X) suspended researcher access to their data. To fulfill the need for bot detection, we revived and rebranded it. Check it out at
1/3
4
38
81
0
4
16
The papers accepted by
#CySoc
2024 are now available online, congratulations to all the authors 👏!
We have a great program for everyone, details at 🚀. We look forward to seeing everyone soon!
If you can't make it in person, DM me for Zoom link!
1
7
16
A common strategy for increasing exposure is to reply to other accounts' posts. Sometimes they choose random posts, sometimes they target high-profile accounts and popular posts. Another strategy is, of course, to interact with other bots, which leads to anomalies.
2/4
1
3
16
Some collaboration tricks they don't teach in school: If you want something to happen, you have to be very clear about who is responsible, when you want to see the results, and what the expected outcome will be. The more specific, the better.
0
0
17
Quick analysis on bot activities of tweets mentioning 'Elon Musk' Apr 23-26. Two notable patterns
- Twitter&Musk deal news out -> higher volume -> lower bot activities
- Tweets expressing overall positive sentiment began to show higher bot amplification levels after Mon 4pm ET
🧶
1
8
16
Just ran Llama 3 8B on a virtual Linux machine with 8 CPU cores, 30G RAM, and no GPU. It was generating output at a speed slightly slower than GPT-4 🤯. Note that this is a ChatGPT-3.5 level model, which is pretty capable already. It just changed a lot of things.
1
0
16