There’s a great paper by Introna and Hayes which interrogates the algorithms driving the plagiarism prevention ‘service’ Turnitin, which raises exactly this point. They argue that once we understand how that algorithm operates, we can then understand how Turnitin works to define and label particular kinds of students as ‘plagiarists’ – well worth a read. The additional issue they raise is that many of these algorithms are not available to analyse because they are proprietary and therefore not visible.

Ben

Ed, I was also intrigued that you chose correlational wheel type visualizations. I think different people are attracted to different types of visualizations. Some people like pie, or bar, or line charts. I tried Textisbeautiful.net myself, and posted on my blog for my ScanFilmTV MOOC micro-ethnography script. I chose topic clouds for that visualization. It illustrates immediately some salient themes because I am familiar with what I am looking at. For example, ‘Danish’ and ‘Swedish’ referring to films are more prominent than ‘Norwegian.’ That was immediately evident to me, but might not be to other viewers. I did find Textisbeautiful to be more useful than other word picture tools that I have tried, such as Wordle. Your 28 February message above is a nice elaboration that helped me understand the underlying explications of the correlational wheels visuals and themes that I might not have grasped otherwise.

So, my question would be about what you think was lost and gained through representing the community in this way. A more traditional ‘ethnography’ might have generated written field notes, so do the visualisations add something more ‘objective’ or ‘evidence based’ here? And perhaps more generally, how do you view data visualisations – do they provide the new ‘truths’ about social life so often promised? Some of these themes will be great to continue thinking about as we enter the third block on algorithmic culture.

Thanks for the comments and prompts, and apologies for the lack of explanation about how the visualisation tool works!

The basic unit of the visualisation is not actual words but topics. textisbeautiful.net defines topics in a particular way. Unlike simple word counts (how many times does a word appear ), topics represent words or phrases that appear closely and frequently in the text. Topics are determined not by word counts but by the tool’s algorithm. Text is beautiful refers to this algorithm as “our technology” and provides some explanations about the visualizations it generates but not much information about how this algorithm works. At least that’s how I understood this!

According to explanations provided on textisbeautiful.net, topics are then grouped according to themes. (Although as a caveat, I did see three themes that contain only one topic: software bug?) Themes are displayed differently based on the type of visualisation. In word clouds, themes are those topics that appear in the same color; in the correlation wheel, themes are the topics that appear within the same arc. Themes should be viewed collectively: each topic within the theme are logically related. A good example of how themes summarize the text is the theme that contains the topics “group”, “post”, “email”, “send”, “forward”, “subscribe”. Another good one is the theme with the topics “circle”, “time”, “join”, “looking”, “members”, “available”. Both are incomplete lists but they summarise how participants tried to organize themselves into groups, primarily via email.

Text is beautiful provides another logical grouping, one that looks at how often a topic appears when another topic is present. Text is beautiful calls this as co-occurrence. Like a theme, co-occurence is calculated, that is, it is determined by an algorithm. Co-occurence is displayed as the highlighted lines inside the correlation wheels. For example, when the topic “silence” appears, the other topics that appear are “stillness”, “images”, “deep” and “metaphors”.

My claim is that the co-occurence of certain topics provide visual evidence of the relationships that participants have built within the MOOC, and that I have read about in the MOOC discussion forum. First, I divided the posts within the forum into two: posts that were about how to organize groups of participants called coaching circles, and posts that were about participant’s reflections about what occurred within those groups. I noticed that the language and tone of those two kinds of posts were very different from each other. The second type, the reflective ones, were more personal, mentioning feelings and emotions more frequently. In the actual posts, several participants reported even crying and feeling deep gratitude after a coaching session. My problem was that I did not want to quote those posts, and certainly not publicly. The correlation wheels provided an abstract but evocative solution. For example, the topic “today” co-occurs with the topics “clarity” and “community”, alluding to the personal insights that participants have gained and have reported happening within the community.

I realise this is crucial information I have missed. Again, apologies for not including this as part of the artefact.

Would also love to know more about how you created the visualisations.

Could you say a bit more about the final visualisation – ‘introspection’ – as it’s not quite clear to me what is going on here, or how the visual has been generated, and how it has informed your brief analysis? How does the correlation between words indicate ‘deep connection’ between participants? It would be very interesting – either here or in the final assignment – to see more analysis of how these lovely visualisations are generated, and how the textisbeautiful algorithms work on text in a particular way to achieve particular effects.