Using computer based text analytics to study the Austin bomber

The recent wave of bombings in Austin Texas has left the United States with very little understanding of motive and political background of the individual. Early reports suggested that the bomber might be a terrorist inspired either by modern day Islamic radicalism or (possibly) a Unabomber type lone-radical.

The recent wave of bombings in Austin Texas has left the United States with very little understanding of motive and political background of the individual. Early reports suggested that the bomber might be a terrorist inspired either by modern day Islamic radicalism or (possibly) a Unabomber type lone-radical. Currently reports are stating his bombings were random targets, although early reports jumped to the conclusion that he may have been targeting racial minorities. Last week, the bombing ended with the death (by suicide) of the bomber. It has been reported that he left a 25-minute confession tape, detailing how he made the devices, but never mentioning any motive to his crimes. Because of how quick the media is to associate bombing and mass killings with terrorism, the individual was almost immediately labelled a terrorist. However, terrorists are defined by the fact that their actions are politically motivated. Yet, no political motive has been found for the Austin Bomber.

One thing that might shed light on the motives and background of the individual could be what he has left behind. In my work, I use a suite of computational algorithms and analysis tools to recreate the mental schemas of people based on how they speak or write (a schema is a structured network of information that our mind stores about a topic). I typically focus on religious and political extremism. It was recently discovered that the Austin bomber left behind a blog of 6 political writings, dating from 2012. It was virtually unknown until this week (as evidenced by a total lack of comments to the blog dating from before the news revealed him to be the bomber.

I was able to download the texts of the blog and, using the algorithms at my disposal, recreate maps of the concepts that the Austin bomber used in his texts. This creates an interesting map of beliefs (a schema) that we can then analyse to understand his beliefs and the structure of his thought.

One of the first things that stands out is how most of the concept maps appear disjointed and rather sparse. This suggests that, at the time of writing, the ideas were not well integrated into a single narrative or set of ideas on their own. Rather, a number of disjointed points appear to be utilized to construct an argument that does not make many connections.

For example, if we look at the map that results from his blog post outlining why he believes gay marriage should be illegal, you see not one large cluster, but many disjointed ones. (For those who are interested I used a program called ORA to create visualisations and charts).

Network from post addressing gay marriage

His post against free abortions shows a roughly similar pattern, having one larger community of concepts and multiple smaller communities that appear to float around the edges.

Network from post addressing Abortions

However, some of his posts seem far more dense in their conceptual maps. For example, his post on internet piracy is effectively two conceptual clusters presented in the same post. This seems to reflect a more general reflection about internet piracy in one cluster and a more focused cluster on “Kim Dotcom” and his role in the internet piracy debates.

Network from post addressing internet piracy

The most visually cohesive writing that the bomber left behind was his post about the sex offender registry. This writing largely presents a single idea as to why he did not believe that sex offender registries should be utilized. This is, on the whole, anomalous given the disjointedness of the other posts as well as the general conservative morality perpetuated by the author (this is discussed more below).

Network from post addressing sex offender registry

A visualisation of his post about releasing a terrorist. Here it does seem that there is slightly more cohesion in the single large network with larger network surrounding it.

Network from post addressing the release of a terrorist

A visualisation of his post about the death penalty. Again we see a cohesive core with small satellite networks.

Network from post addressing the death penalty

We can also use a similar algorithm for all of the material in the website. Typically, this can be done using an algorithm called a web-spider, but given how few posts there are, that would be a bit much (copy and paste worked just fine). This creates a network of all the concepts he puts together, linking not just concepts within a text, but between texts as well.

Here, we find 6 conceptual clusters (highlighted by different colors below.

Louvain modularity analysis of union network (visualised in ORA)

If we add links to the texts that the concepts are drawn from, we see that these clusters roughly correspond to the individual posts he put on the website.

Louvain modularity analysis of union network with texts links and nodes (visualised in ORA)

This also suggests that there was not a single cohesive narrative or set of principle beliefs that were applied to different political issues. Rather, there were smaller points, related only to a specific social issue that the post focused on, that were used at any one time.

This is interesting as it suggests that the bomber was not a Unabomber type individual who was highly committed to a specific ideology or belief system at the time of writing. This is a point I come back to again in a second.

First, mining these networks also allow us to look into what kind of concepts are most central to his belief system. We can do this a couple of ways. For example, we can look to see which concepts are most frequent:

Distinct frequency plot (generated in ORA)

We can also see which concepts have the most links to other concepts (called total degree centrality). This is an interesting approximation in my experience because it begins to show you what concepts are the most important to a set of beliefs. Here we see that the term “not” is most central. This is not unconventional for conservatively leaning texts in my experience.

Centrality plot for all texts (generated in ORA)

In addition to a network based analysis, we can also use algorithms that find how much of a text is addressed to specific themes by using word frequency counts. This can reveal how negative or positive the words are, or what kind of underlying moral foundations the individual had — if any.

Analysis of writings from bomber’s blog. Presented by theme.

What this reveals is that — surprisingly, the posts are not filled with high levels of negative emotion, sadness, or anxiety. Even more telling, many researchers are currently investigating how fundamentalist groups throughout the world who become “fused” to a social or religious ideology use a surprising amount of family-based language. In his posts, the bomber used none.

Lastly, research has established that different political ideologies in the US have different moral “signatures,” or sets of moral themes that are more stressed than others. For example, Liberals focus on harm and fairness more than conservatives while conservatives focus on the in-group, purity, and authority more than liberals.

In the posts he left behind, the bomber’s moral signature appears to be somewhat ambiguous. While he is a self-identified conservative (according to the profile he wrote for himself), harm is the most prominent category in his writings. This is more in line with liberal moral patterns than conservatives. Yet, fairness is almost entirely neglected in his writing. With the exception of moral language dealing with harm, he does seem to use authority, purity, and in-group language more frequently.

Analysis of blog for moral language

From the analysis here, we are left with an interesting picture. I interpret this data as reflective of a highly conservative individual, but not an extremist. Yet, something clearly happened that resulted in his undertaking of a bombing campaign. This data comes from 2012, so to my conclusion is that something happened after 2012 that either 1) resulted in a serious mental break or the development of some psychopathology or 2) radicalised him to a hard right wing viewpoint.

Recently reports have surfaced that the bomber left a confession tape, but that it leaves little information in the way of a motive. I find this suggestive that option 2 above may not be the most likely. Typically extreme radicals use their deaths to support their cause, and in doing so, they can use it as an example to get media time to promote their beliefs. Individuals like Timothy McVeigh, Ted Kaczynski (Unabomber), and David Koresh all had a great deal of things to say prior to their deaths that were supportive of their ideology.

At the moment, this looks to me as someone who was both a conservative (who leaned further right than most conservatives in the US, but is not — as seen in these writings — “alt-right”) and somewhat confused. He did not seem to have a clear set of beliefs. This raises the possibility in my mind that he did not have a clear cut social identity that he could attach to, as such he likely had few close friends or any one group that he felt at home with. I would bet that he became increasingly isolated in the years following the publication of his blog and, at some point that I cannot discern here, an event — or series of events — led to him committing these heinous acts.

For those who are interested I’ve uploaded all the texts, pictures, and network files to github here: