Hate Speech decreases finally?

Abstract

No, there was no decrease of hate speech in the German social media buzz during the last 6 months. As already seen during the last 2 years, the largest number of hate speech is found in twitter (70%), followed by facebook (10%), blogs and forums.

I have developped a detector which finds public hate speech posts in these pagetypes. Read here, how this detector works.

Findings

Here are the data the detector produced for all 4 page types from May to October 2017:

The peaks mark terroristic attacks during this summer. We had a knife attack Jul 28th in Germany and the Barcelona-car on August 20th.

Due to the dominant role of twitter in the diagram above, Facebook’s performance is somewhat flattened. Here is the diagram only for Facebook:

A decreasing line looks different.

Now, before you interpret and use some of the numbers shown in the diagrams, you should read however the following paragraph.

Method

The detector works very simple, using a common social media monitoring tool. The queries for the detection-research look for posts containing “hate-words”. Hate-words are swearwords which mean people give to others, who they hate, to muslims, blacks, jews, foreigners and so on. You know some of these words of course, I don’t have to list them here. They are all ugly.

Now when a post contains such a hate-word, the post is counted as hate speech.

Does that cover all posts in question correctly? No, it does not.

  • The detector finds post which are no hate speech. For instance if I had given an example of a hate word in this post, this very post would count. It contains one of the ugly words.
  • The detector does not find true hate speech. The phrase: “One should really hang all xyz!” contains no hate word (as long xyz itself isn’t one). Hence, when looking through the detector’s lens, this phrase is no hate speech.

So the real volume of hate speech may very well be quite off the number we have found with the little detector here.

But how relevant is this absolute volume? Much more important to me is, how this number develops. It is quite obvious that its curve will have a very similar shape as the one of the detector. When the true hate speech volume increases the detector will show an increase too – and vice versa.

More information

I have separated the queries by “peer-group”. One query f.i. is called “jew-enemy”, another one “muslim-enemy”. The curves for these queries are not completely parallel and hence provide further interesting insights.

Since the detector now runs since more than two years, I have collected some data from that time. It is avalaible in principal.

Contact me if some of these data are of any interest.

 

Hatespeech increases after Terror in Paris

Some weeks ago we’ve set up some hatespeech queries using a common social media monitoring tool. We focused only on German posts and mentions. To use the appropriate search strings requires quite intimate cultural insights. We did not dare to stumble through a foreign language on a subject so sensible as this one.

Hence this is only a brief English version of the corresponding German article about our findings. The full German version is quite a bit longer and it contains diagrams. To read it please follow the flag in the top.

The Setup

We focused on the sources Facebook, Twitter, Blogs and Forums.

We had 7 queries to catch hatespeeches against asylum seekers, blacks, Jews, Moslems, Roma people, Turks and foreigners (“Ausländer”) in general.

We searched by typical hatespeech words such as “Kümmeltürke”, “Judensau”, “Nigger” and the like.

Some findings

Over the last 4 weeks we found:

  • There are 800 – 900 hatespeech-posts per week. The absolute number however is by no means representative. Every new search string produces new results.
  • Among the results are about 45% posts which are no real hatespeeches. Some Turks for instance call themselves “Kümmeltürke” as some kind of a joke. Quotes of hatespeeches are among our findings too. And articles like this one as well. These and a some other types do not represent true hatespeeches.
  • The majority of the hatespeeches is on Twitter, not on Facebook. Actually the Twitter volume is nearly three times as big as the Facebook volume. But we have only about 1 million Twitter users in Germany compared to 30 million Facebook users.
  • The majority of the hatespeeches is against Moslems, Roma people are target no. 2. Turks and asyl seekers are rather a minor target.
  • The attacks of November 13th in Paris brought a sharp increase of hatespeeches – but only against Moslems.

Next Steps

We plan to go on for a while with this subject and produce some more results. Especially the ratio true vs. false hatespeech seems worth to be considered in more detail.

There is strange meta finding with this issue however: Nobody wants to be quoted or referred to a topic like this. Of course: We use quite a lot of really bad words. They are far from political correctness. Some machine-like checkpoints and control-bots might detect this and put the article and everything in relation to it on an index – bad reputation management.

So maybe we are soon turned off by the provider of the monitoring tool or even by somebody else – we’ll see. Stay tuned!

And as always: any serious comment on this is absolutely welcome!

Social Networks: Users in Germany and Worldwide

In this article I will provide

  • an overview over the number of users of the 10 most relevant social networks in Germany in 2014,
  • a comparison of these data with the corresponding worldwide figures and
  • a breakdown of the German data by sex and by age.

Read more