Week 4: CCK11 & Creative Potential

Interpretation of Creative Potential in Network Dynamics across Social Media

Abstract

Analyzing network data to uncover correlations involves what you expect: clear understanding of network measure definitions and thorough knowledge of your subjects and their context. There are two challenges. The first and most obvious is the combinatorial complexity. Second, there must be some work bridging the gap between abstract network measures and the more intuitive concrete behavioral narratives. In other words speaking of only learning activities and network measures is inadequate to these two tasks. In this article I suggest performing this matching activity using tables by performing the following steps, Define – Describe – Relate:

  • Define network measures
  • Describe basic Twitter and blog activities/behaviors/actor goals
  • Describe Learning contexts
  • Relate these three in a table while noting their Relations: symmetric, asymmetric.
  • Interpret Gephi and Tableau visualizations using the table to bridge the measure–behavior gap

Only after performing this groundwork can we look for novel interpretations in SNA.

Introduction

The trained network analyst may respond to this article by saying the principals discussed here are simply those of learning itself and not specific to network analysis. To this I respond by saying this article tries performing two tasks at once. The first is indicated by the title given in the course assignment. The second is derived from the cognitive dissonance between narrow mathematical definitions and human behaviors, goals and interactions.

I don’t personally work within academia, therefore, if there is an excess of description relating human behaviors to user experience patterns to the detriment of network measures, its because I’m working without proper context. Thus, as a writer I need to establish this context to begin writing. As a reader, you may appreciate the context for my Conclusions.

 


Figure 1-1: Twitter Network Betweenness Centrality ▾

cck11_twitter-3.1a-t12


Figure 1-2: Twitter Network Betweenness Centrality with degree filter (min 2 nodes) ▾

cck11_twitter-3.2e-t12

Nodes over the Whole Network

Comparing the first two diagrams, you can clearly see the change from figure 1-1 to figure 1-2. Low-volume blog and tweet users (< 2) simply fall off the graph. Manipulations such as community detection are the first analyses performed at the macro level of the whole network.

At the other extreme we are working to discover the correlation of network measures and our topic of interest–in this case student ‘creative potential’. Now, we change the focus of our analysis to the individual cases of student communications.

Measuring the potential of particular nodes

Below are two tables containing permutations of possible scenarios between students in the CCK network. These Actor Tables list logical student activities, reasonable learning contexts and possible network measures. An additional column, labeled “T” attempts to analyze logical consequence between these categories.

For example on line 3 of table 1-1, “Student B up-votes Student A’s post; Student B’s objective is to exercise their ‘vote speech’.” Student B’s learning context is “Participating in cooperative learning (by providing feedback) and recognizing a quality answer or acknowledging a shared interest.” This relationship is symmetrical. Which is to say, if a student up-votes a post, you may assume they are ‘recognizing quality‘ or ‘expressing shared interest‘. And if a student’s intention is to ‘express shared interest‘ the student could be expected to ‘up-vote‘. A implies B, B implies A. Finally, this relation is marked as symmetric with the network measure Betweenness Centrality which we are working to associate with Student Behaviors.

This is the important logical step in developing the Actor Table as a tool. Can these series of behaviors be understood as a symmetrical, asymmetrical or non-relational? The goal in this step is to develop a set of possible cases, logical narratives, which can each be proposed in-turn as one of these three possible relations. Thus the interpretation of abstract quantitative network measures is reduced to the combinatorics of network measure definitions, technological interactions, and our learning contexts.

Before we go on, let me be clear, the CCK data set does not include voting data. I’ve added it here only as part of the conceptualizing process necessary to work through this researcher’s knowledge and experience with social media. Post voting is the simplest form of comment.  A typical blog comment could contain any potential message (i.e. “You misspelld ‘critical'”). Post voting is rather like an archetypal voting scenario. Whatever the commenter’s original intension, the message is converted to a quantitative value. (It’s such a strong paradigm for me, I just had to include it to eliminate the confusion of its absence).


Table 1-1. Communication Context.

Student A writes Blog Post In Learning Context. Student B responds.
IDNet. M.TActorActionWishTLearning Context
01na'Awrites a blog post share wide audiencedemonstrate received knowledge
02na'Atweets regarding new postpublicizeparticipate in cooperative learning
03*BCB**up-votes A's postvote speech: recognize quality or shared interest participate in community
04BCBcomments on A's postdeclaring presencedemonstrate received knowledge
05na'B**bookmarks A's postcollect informationrecognition of quality or shared interest


Table 1-2. Learning Contexts between individuals

Students B & C respond to A's Blog or Tweet
IDNet M.TActorActionWishTLearning Context
06BCBtweet A's blog or retweet A's tweetact as brokerinfluence of A→B
07BCCcomment on B's retweetdeclare presenceinfluence of B→C
08*BCC**up-vote blogvote speechinfluence of A OR B→C
09BCBComment on A's tweet or blog declare presenceinfluence of A→B
10BCB^ Follow Ameet need (social, information, etc)influence of A→B
11na'X**bookmarkmeet need (information)influence of A→B


Key

Net.M.Network Measure
BCBetweenness Centrality
(*)IFF blog site requires unique ID or student community is of a size where a nick name can be reasonably associated with one classmate.
(')No Network Feedback.
(^)Follow of course is a one time action.
(**)These data are not in CCK11 data set
All examples assume honest intentions on the part of students.

Again, these centrality scores are related to messages, sent or received, regardless of topic. However, in the learning analytics we benefit with a highly motivated context, which is to say students are apt to communicate about learning, as opposed to their social lives in above examples 6-9, and 11. However, in case 10, a Follow action could represent an exo-learning interest (for example an out of school friendship).

Interpreting twitter & blog networks in terms of Creativity

Network analysis principles define betweenness centrality as “Number of shortest paths that pass through a particular node.” [4] And, again from Prof. Gavisic explained, betweenness centrality is a measure of the ease of connection with anyone else in the network.[5] Actors with high betweenness centrality we are calling Brokers:

“…early diffusion research identified as an opinion leader, a person responsible for the spread of new ideas and behaviors.” [6]

In the context of Twitter’s message content and interactive features, Brokers are exposed to more–more content and more people. Surplus is frequently associated with creative solutions and is very common in the arts. While we commonly think of prose writing as an iterative process (continual improvement), who hasn’t written something clever which, nevertheless, doesn’t quite fit in the flow of what you’re writing. In the arts, and in poetic writing, each scrap or version or hint or glimmer of an idea–never more so than when we’re excited by our own cleverness–is preserved for some future reference or use. (See for yourself. Visit the Flickr groups Moleskinerie (at 130616) or One Page at a Time (at 62619) for examples of artist sketch books).

Brokering is not symmetrical in the blog context

In the twitter culture there seems to be a free flow of links and ideas. In fact, the user interface includes this capability in the experience as retweeting. The blog culture is very much different, and follows more the pattern of traditional publishing’s author and reader relationship. When you comment on someone’s blog, you’re acknowledging, as in the example above, a quality in the work you admire. Or, possibly you wish to assert your presence within a social community. Whichever the case, unless the commenter is also blogging, they are assuming the role of consumer of an author’s creative production.

These are two different forms of social capital, and both relate directly to the form of the expression. A tweet is a link or image or brief text. It’s the punch line without the joke. It’s the hot fudge, without the big bowl of ice cream. While a blog is that bowl of ice cream, and it’s the author who gave it to you. The former, while delivering content, is equally a social interaction. Twitter operates on the edge of timely information with a thin veneer of name dropping your source. In other words the highest form of capital in twitter culture is to be eye-witness to an event and have your message retweeted again and again. Your network of followers is your potential for creativity. This is participatory creativity: a tweet is your expression. A tweeter’s creative dividend is the potential timely knowledge gained during unusual circumstances or an increasing number of followers.

A blog’s principal main focus is its message: a text, image, or other digital work. For the blogger the delivery of personal and original content is the accomplishment. Self-expression is the creativity–everyone else is just lurking with the possibility of feedback. The productivity of a blog is limited only by the possible output in any 24 hour period. Quality is judged one piece at a time. The commenter is always at some distance from the material of the blog post, whereas a Twitter user seems to be part of the content by their act of tweeting/retweeting messages into their corner of the Internet. Are you acting creatively in the act of commenting? No. The author is the creative actor. The commenter is spending their time not writing.

What’s the point of this stream of consciousness assessment of blogs V. tweets? Until we, as analysts, are able to produce a sufficient number of social network communications correlated with student behaviors and learning contexts, we may struggle to fill the gap in our understanding between the extremes of whole network analysis and single actor analysis. For example, what is the ratio of a blogger and their frequent commenters compared to tweets of any one particular blog post. Or, to put in other terms, what is the ratio of followers to novelty seekers in a blogger’s network? What is the retweetability of a blogger’s or tweeter’s content–how far does one author’s output spread on average? Which is to say some posts, while expressing quality of information may simply generate the dreaded reaction: TL;DR in the dominant social circles on social media.

Developing an intuitive model of network measures.

There are two immediate objectives for network analysis for the introductory student. First, it’s important to begin to develop clear understanding of what the available measures signify. The intro quote above suggests ‘centrality’ associates with ‘influence’. How do we get this notion? It comes from the significance of comparing multiple centrality values. When a node has low centrality, few or none nodes must cross this node to reach any other node on the same network, and high value reflects the crossing of many nodes. It comes from our through understanding of the relationships inherent in the platforms of communication in our data.

Second, we need to grasp this quality of network measures which measures the potential of each node compared to every other node within the same network. Take an the counter example the famous mathematics word problem, the Seven Bridges of Konigsberg[2], which asks: What is the path you must take to cross each of the seven bridges only once and return to where you started[3]? The Konigsberg bridge problem is only “one path”, while our centrality measure is for every node in the network. It’s important we understand the possible characteristics of these relations, understand the possible narratives as best we can (there’s really no way of knowing all the ways people use technology). Fortunately, the learning context limits some of the more troublesome possibilities such as political interest, coercion and falsity.

Conclusion


I’m most excited by figures 2-1 and 2-2. Figure 2-1 shows indications of increasing betweenness centrality in twitter at the end of the CCK course. This might be an indicator of Brokering behaviors.

If brokering is positively correlated with high levels of creativity, we might expect to see more different individuals taking up blogging also, because they were inspired by tweets or blogs. But that’s not what figure 2-2 indicates. Instead we see the same bloggers continuing to blog, while the overall tweeting volume increases. What’s more, of the six high-frequency bloggers, four of them are also medium to high volume tweeters. What exactly are the tweeters talking about? The course content? The six high scoring betweenness and degree bloggers’ content? From the data we can’t tell. Looking only at the network data, bloggers appear as the more creative individuals using both platforms. Tweeters are happy and content tweeting the time away talking about…? Whatever they can squeeze into 140 characters–obviously!

Further investigation could match each high-frequency blogger’s Unique ID to their individual commenters. Are these the same users or are they different? Do these patterns persist with all the bloggers (are there cliques), is there diversity, or are comments more equally distributed? Tracing the blogger IDs to their twitter accounts and further one level to connected twitter users, what ratio of twitter followers are also blog followers? These are some of the interesting questions which occupy the middle ground between whole network analysis and individual nodes [7].


Figure 2-1: Blog & Twitter Degree v. Betweenness Centrality joined on node ID ▾


Figure 2-2: Blog & Twitter  Degree v. Betweenness Centrality joined on Label ▾

NOTE: There are some irregularities between the last two columns. Each table is ordered in color pairs (t6, t12) where the latter data set should include the former. Notice these last two columns appear swapped; however, Tableau shows the header Measures are in the correct order–save for one detail. Its difficult to describe, but Tableau appears to have combined (t6,t12) in column seven. I have a forum post seeking assistance on Tableau.com and will update this section when new information becomes available.

UPDATE_12-01-2014:
One mistake (mine) and a mystery (Tableau) conspired to convince me columns in a key graphic were swapped. However, I misinterpreted Betweenness Centrality as cumulative value similar to degree. A weird join issue in Tableau, having to do with taking data from four external .csv files, reinforced this misinterpretation.

Network degree (the number of connections) in our data set accumulates over time. In other words blog posts at week 6 are a subset of blog posts at week 12. However, Betweenness is not cumulative–it’s a measure. Ans some individual’s Betweenness went down and some up by the end of the 12 weeks.

Tableau does something automatically, which I still don’t understand, when I construct the graphic. It combines in the column-pill’s rollover the names for two files: t6 & t12!. I recreated my data set then and found the same result. I recreated the data set again today and the weird combination switched from Twitter to Blog, while the graph remained the same. That’s when I realized the true nature of Betweenness in this time series comparison–it’s a score, not a count. The quick check I used to judge degree (`t6 <= t12`) did not apply to Betweenness. Funny thing was, Blog Betweenness did increase. It was only Twitter Betweenness which had a few noticeable declines. The Tableau question still remains a mystery to me, but an important detail of SNA has been revealed to me by a mistake.


See also the Network Details for the above figures.:
Week 3, CCK11 Twitter and Blog data network import settings & measures


Related posts:
Opinion: Creativity & Data Science


Footnotes

[1]: Shane Dawson, Jennifer Pei Ling Tan, Erica McWilliam. (Australasian Journal of Educational Technology, 2011, 27(6)) “Measuring creative potential: Using social network analysis to monitor a learners’ creative capacity”. p930

[2]: Seven Bridges of Konigsberg Problem [Wikipedia]

[3]: No such path exists.

[4]: Hirst, T. (05-10-2010). Getting Started With Gephi Network Visualisation App – My Facebook Network, Part III: Ego Filters and Simple Network Stats, (Retrieved October 18, 2014)

[5]: DALMOOC Week 3, lecture 3: ‘Network Measures’

[6]: Ronald S. Burt, Martin Kilduff, Stefano Tasselli (08-2012) SOCIAL NETWORK ANALYSIS: FOUNDATIONS AND FRONTIERS ON ADVANTAGE p.7 [Via DALMOOC Resources]

[7]: Writing this piece I worked to match certain language patterns from the lectures with my use of terms. One lexeme I have not yet wrapped my head around is ‘potential’, as in ‘node potential for creativity’. It must be a learning only context, because I don’t associate it with communications proper, artistic creativity, design, or programming. The nearest point of reference for me comes from the scientific domain’s concept of ‘potential energy’. That possible amount of energy which could be produced if only the initializing reaction is achieved.