Luca Hammer Profile picture
Data analyst. Working on a tool to evaluate Twitter accounts: @accountanalysis. 📜 Code of Conduct:…
Konrad Paumann Profile picture 1 added to My Authors
25 May 20
Which combinations trigger the COVID-19 info label? 5G plus a Corona term for example. But adding “fak_enews" (no underline), removes it again.
5G Corona
5G Corona fakenews
Read 16 tweets
30 Apr 20
Free startup idea: Online course for celebrities how to monetize their telegram channels full of gullible fans that neither trust science nor media. /s
- Every bad review or investigation is an attack against you and only proves that you are right.
- Sell any product by claiming it helps against whatever you tell them to fear.
- Ask people to just give you money, because your fight against the evil powers is expensive. /s
- Allow them to invest in a made up project to get more money in less time. Sadly it doesn't work out and all money is gone. Give them equity in the next project to make them whole again, but they will have to double their investment. Repeat.
- Cross-pollinate other channels. /s
Read 4 tweets
11 Apr 20
Let the markets do their thing. Give the money to the people and it will trickle up to the companies who provide the people with the things and services they need.
Austria plans to give a billion euros to an airline. Instead it could give each employee 100k€. Let the employees decide, if they want to rescue their airline or do something else. Rescue the planet for example.

Minister president of Bavaria suggests rewards for car makers if the make modern cars. How about laws that disallow them to make awful cars and let the people decide what they need?

Read 7 tweets
13 Mar 20
According to a simulation by TU Wien and dwh, reducing social interactions by 25% would reduce the maximum of concurrent Covid-19 cases from 2M to 360k in Austria.…… without measurements to reduce social contacts: 2 million infected at end of may.<br />
with 10% reduction of social contacts: 1.2 million<br />
with 25% reduction: 360 thousand infected in summer
People who found this research interesting also followed @niki_popper.
Austria hast 44k regular hospital beds and 2,5k intensive care unit (ICU) beds. There are no official numbers, but based on past occupancy, around 80% are currently in use.…
Read 5 tweets
7 Jan 20
There were some questions about the data analysis I did for this Spiegel article:….

I collected them and try to answer them in this thread. /0
I used the Twitter Standard API to collect the Tweets. The full query was: “OmaUmweltsau OR umweltsau OR nazisau OR wdr OR wdr2 OR UmweltsauOma OR UmweltsauGate OR k2912 OR TomBuhrow OR Buhrow”.

Time frame 2019-12-27 6am to 2019-12-30 6am. /1
Including "wdr" and "wdr2" increased the number of false positives. After looking at Tweets that wouldn't be in the dataset, I came to the conclusion that including them creates a more complete picture.

Comparison to Tweets before the outrage showed low background noise. /2
Read 8 tweets
16 Nov 19
German first aid kit against social media attacks

KĂĽmmern sich um das Stellen von Strafanzeigen.

Beratung und Finanzierung von Zivilklagen.

Beratung und UnterstĂĽtzung in Ă–sterreich.

Entschärfen durch Counterspeech und Solidarität.
@hassmelden @HateAid @Verein_ZARA @freiheitsrechte Weitere Projekte:

Dokumentation antisemitischer Vorfälle und Unterstützung für Betroffene.

Verband der Beratungsstellen fĂĽr Betroffene rechter, rassistischer und antisemitischer Gewalt.
Dazu gehören
Read 10 tweets
6 Nov 19
I wanted to push my latest two tutorials to Medium because my older articles got some views there. I stopped using it as the first place to publish something, when they introduced their signup wall.

They turned off the WordPress-to-Medium plugin, so I had to use the manual import. Which worked okay.

But then they told me, that I have to enable the metered paywall for may articles, if I want them to be distributed to more people than only my followers.
The articles are freely available on my website. I am okay when people pay for them on Medium. I activated the metered paywall. How do I get paid?

Per default I don't get paid. I have to join their Partner Program first. To do that I need a Stripe account. Luckily I have one.
Read 5 tweets
5 Nov 19
A student asked me how to collect old Tweets with the Twitter Premium API, so I wrote a tutorial how to get started with the Twitter Premium API.

Looking forward to your feedback.… #python #twitter #datascience
Because the collection of Tweets is only the first step, I shared my script that creates a dynamic Retweet network from the jsonl-file.

Have fun.… #gephi #gexf #python
What do you call Lehrgeld in English?

Read 4 tweets
31 Oct 19
Insights into a polish astroturfing company. Journalist Katarzyna Pruszkiewicz worked there undercover for six months. 14 people managed about 200 fake profiles to push the agendas of their clients.

Polish article:…
English will be published by @guardian.
Depending on how well they are executed, such operations are nearly impossible to identify from outside. They are more dangerous than automated spam (aka bots).

The article shows how such operations are often embedded into other marketing activities. /1
Their clients got impact reports (how many accounts talked about the company, product, issue; what they said) but they hid the fact that most of that content was fabricated by themselves.

Everyone is happy while they destroy public discourse. /2
Read 9 tweets
3 Sep 19
ML: This account has a probability of 30% to be a bot.

Me: The Account is active 24/7. It tweets at least two times per hour. Most of the tweets are posted through an app that nobody else uses. It claims to be a bot.

But it's not fully automated and something is suspicious. /1
When you only look at the Tweets through the allegedly bot interface, there is a wake-sleep pattern in addition to the 24/7 activity. Obviously in a different time zone than me. Probably America.

That pattern matches the wake-sleep pattern of the account when used manually. /2
I assume that the automated interface is manually fed or additionally triggered. Could be for testing, could be to disguise human work as the one of a bot. /3
Read 6 tweets
5 Jul 19
An account that can't be blocked. Interesting. #twitter #bug
After reloading the account is unblocked again. Other actions don't work either. Following increases the follow count but the account isn't followed.
Little warning: Trying to follow the account increases your own following number as well. (And you can't reduce it, because you can't unfollow the account as you are not following it.)
Read 6 tweets
17 May 19
Tweets started at 18:00 MEZ (16:00 UTC) at 10 Tweets per minute and rose to 30 Tweets per minute.

Read 14 tweets
5 May 19
Twitter is has locked several accounts in Germany because of misleading content about elections.

Most prominent examples: @RAStadler
and @SawsanChebli.

In this thread I try to show different perspectives, because reality is complex. /1
Misleading content about elections isn't a new rule by Twitter.

Because the Twitter rules state that users must adhere to local laws and there is a German law against misleading voters it always was prohibited.…… /2
What changed is the option to report such content.

It was announced on 24. April and rolled out in the EU on 29. April. /3

Read 14 tweets
26 Feb 19
Today a 21-pages document was sent to at least 22 members of the European parliament. The authors claim to „have clear evidence of US meddling in this EU law making process“ and that a full release will happen in the next days.

They are wrong and here is why. /1
Good: They are transparent about how they came to their conclusion and what tools they used. Else this thread would have ended like the last one.

The false core claim: 23%-47% of mentions of anti copyright directive campaign hashtag mentions came from the US. /2
Biggest mistake: They presented location data by @Talkwalker without questioning it.

The used the full version, but the free version is enough to prove my point that the location data is not trustworthy.

Let's look at #SaveYourInternet as an example. 63k mentions in 7 days. /3
Read 24 tweets
6 Jan 19
0rbit released the illegally copied data over the course of a month. Why did nobody notice it?

The followers of the account were all from one cluster and most of them weren't active in months.

#hackerangriff #datenklau #BTleaks /1
_0rbit had around 18,5k followers when Twitter suspended the account (). Thanks to socialblade we know that it had 16 580 followers on the day before the @unge hack. Based on that I was able to reconstruct the followers. /2 screenshot socialblade
We don't know when the followers followed the account, but we know when their accounts were created. Until 2017 it could be organic (popularity of Twitter in Germany). But in April 2017 there is an unnatural increase.

The follower numbers show same pattern. /3 graph when accounts were createdgraph how many followers the account had in the past each month
Read 6 tweets
10 Dec 18
This paywalled article claims 28% of all Tweets about #migrationspakt were posted by bots and therefore the debate was significantly influenced by bots. I don't believe that and here ist why.

(Thanks @Heimatlost for bringing it to my attention.)… /1
- I was contacted by the journalists about their investigation but never got an answer to my reply (attached).
- I was approached by Botswatch in Person in 2017 and did a video call with their team to explain my approach (unpaid).

The article doesn't explain how the bots were identified. Therefore it's impossible to confirm or falsify the claim.

I know of three approaches:
- Machine learning (botometer, botrnot, etc.). Features of accounts are compared to features of known bots (learning set). /3 Botswatch ist ein „Cyber Intelligence Startup“ aus Berlin, das eine Technologie entwickelt hat, um die Beeinflussung der öffentlichen Meinung auf digitalen Plattformen durch solche Social Bots darstellen zu können.
Read 19 tweets
17 Oct 18
Twitter released multiple datasets with Tweets by Russian (IRA) and Iranian accounts.…

I just downloaded the Tweets (…) and will add my findings in this thread. My focus is Europe and Germany.

/1 Screenshot of a Python jupyter notebook, showing that there are 9041308 Tweets in the IRA dataset.
This is work in progress. I make errors. There is no editor who checks my Tweets. Feel free to ask questions. /2
@DFRLab got the data upfront.
Thread and article….

Their conclusion: Little to no impact. /3
Read 83 tweets
14 Aug 18
I love feeds and the people I follow on Twitter.

To make it easier to add the websites of the accounts I follow to @feedly, I wrote a little tool that looks for feeds in your URLs and puts them in an OPML file.

Try it with your followings: Screenshot of the Twitter Followings to OPML website.
1. Login with your Twitter account
2. Keep the tab open until the download starts*
3. Import the OPML-file with your favorite feed reader
4. Enjoy unfiltered content

* It takes a lot of time, because the tool has to visit every website to search for the feed.
Bonus step: Unfollow feeds that publish too much content you aren't interested in.

Choose the "Twitter Followings" category and sort by stories per month.
Read 12 tweets
28 May 18
There is a new Twitter signup/onboarding process.
What it looked like in 2017:
What it looked like in early 2018:
Auto-locking of accounts:

Per default they want a phone number instead of mail.
Instead of a full page signup flow, it is now a modal window.

First screen after entering Name and phone/mail looks like #GDPR. Everything opt-in as the law requires.

I don't know why I get the same screen again. But there is now a button to actually sign up.

Next they want me to enter a password. But the account is already set up (see name and empty avatar at the top right).
Read 15 tweets