lazykoblog   occasional musings about knowledge organization

I suspect in the year 2010 the bandwagon is so old a metaphor few remember what they actually were like. Originally it was a wagon that carried a band in a parade. All of the recent meanings of the term derive from this, because the music is the fun thing in the parade that makes spectators want to follow along or climb aboard. Oh well.

As I read study after study of social tagging I began to wonder about the behavior of taggers. Most studies have demonstrated various properties of the tags themselves, and several studies have suggested tagging is some sort of egalitarian indexing-for-the-masses that would be ever so more useful if the taggers would just stick to a thesaurus. But I considered both of those assumptions unlikely. For one thing, if you inhabit a social networking site just enough to watch the tags go by out of the corner of your eye each day you see a surprising number of them that are self-centered expressions (not just “todo” though there is plenty of that, but also “wtf” and so forth). Also, again watching out of the corner of your eye, the really fascinating thing about the tags is the network of associations among them–in other words, what happens if you click on one, and then when you get to that destination click on the first one there, and so on–you’ll not be following any road that a thesaurus would have led you along (stay tuned for a blog entry about my work at VKS with Wikipedia). There was a lot of discussion about the difference between the main tags and the little ones populating the outer corners of those tag clouds as well, and that reminded me of the problem of noesis, which is the ego-act of perceiving through one’s own experience–this is a hallmark of Husserl’s phenomenology.

I designed a study of tags as exploratory, with the purpose of surveying the tags assigned to a random sample of sites in Delicious.com. I wanted to compare what I would find to prior studies to see whether there was any theoretical potential (there was), and then subsequently to analyze the behavior of the taggers to look for noietic behavior. I submitted an abstract to this effect to the 11th International ISKO Conference in Rome, and also I drew my sample all in one day. I based a sample-size calculation on prior studies’ figures about the proportion of affective tags, and then in my enthusiasm drew twice as many cases (sites) as I needed for 95% confidence. I was excited to get my feet wet with this kind of research. I’m glad I drew the sample manually so I could watch the data as I downloaded the sites and their taggers and their tags. But now I know why people use crawlers for this! My abstract was accepted, and along with it came some helpful referee comments, which sent me to the literature of cognitive linguistics. Bear with me, I was on a learning curve here.

For the conference in Rome I wrote a summary paper about the behavior I observed among the taggers. I discovered plenty of noietic behavior, and interestingly enough, although I was able to affirm the proportion of affective tags–the figure from my study fell within the confidence interval of the prediction from prior studies–the surprise was that the noietic tagging was not affective tagging. I also analyzed the entire sample to see what I could learn about co-tagging–in other words, which taggers were tagging together, and here was my first surprise. A substantial core of the taggers were, in fact, all focused on work on the same sites, and their co-tagging was nested in two clusters, which I was able to identify roughly as web designers and programmers (remember, we’re talking about Delicious.com); the web designers’ tags were descriptive and the programmers’ tags were slightly more likely to be affective.

All of this convinced me I had figured it backwards–the noietic behavior was not the weird stuff in the long tail, but rather was the common ego-act perceptions of the tightly-knit group of co-taggers. In other words, here was a group of taggers all leaping on a bandwagon and in so doing classifying their commonly tagged sites with some very specific and (for taggers) relatively precise terminology.Here is a slide from the PowerPoint presentation of that paper. On the left you see the clusters of taggers, and on the right their tags. The point was that most of those tags could be seen as semantically related to two conceptual clusters–noesis as bandwagon effect. The paper is available in the ISKO Proceedings of the conference at Rome (Richard P. Smiraglia “Perception, Knowledge Organization, and Noetic Affective Social Tagging” pp. 64-7) but here is the abstract:

Knowledge organization can be postulated as existing on a continuum between classificatory activity and perception. Studying perception and its role in the identification of concepts is critical for the advancement of knowledge organization. The purpose of this research is to advance our understanding of the role of perception in knowledge organization systems. We briefly review the role of perception in knowledge organization and some preliminary evidence about affective social tagging, which is seen as a form of everyday classification. We consider how Husserlian phenomenology might be useful for analyzing the role of perception in affective social tagging. Finally, preliminary results of an empirical study are reported.

Because this was for ISKO I was intentionally focussed on the KO issue, which I here stated as a continuum between classificatory activity and perception. I gave a paper on noesis at ISKO in Montréal as well (scroll down, it’s the mailbox paper). I think that we think too often that classification is about putting things in little boxes, and therefore that we think too little about how fuzzy are the boundaries of those boxes. So here is just a glmpse at that issue.

As I said, the referees had sent me to cognitive linguistics, and I found particular resonance in the writing of Ronald Langacker (Langacker, Ronald W. 2005. Dynamicity, fictivity, and scanning: the imaginative basis of logic and linguistic meaning. In Pecher, Diane and Rolf A. Zwaan eds., Grounding cognition : the role of perception and action in memory, language, and thinking. Cambridge : Cambridge Univ. Pr., pp. 164-97). Scanning is the linguistic activity in which a kind of shorthand is used to project a landscape on which perceived activity is taking place; it results in “fictive” or at least unfactual language, but common understanding allows and even encourages this. Here’s a PowerPoint slide from my presentation at CAIS in Montreal in June.

The example is the phrase “my teacher’s books keep getting longer.” What is meant is that each time the teacher writes a book (or, one supposes, even buys a book) it is longer than the last. But that isn’t what was said at all, and obviously the idea that the teacher has a stack of books that is somehow stretching is absurd. It seemed likely that some of the variation in tagging might be due to scanning.

I wanted to complete the statistical analysis of the data and to present a fuller account of the study apart from the philosophical issue of noesis, so I submitted an abstract to CAIS for this year (2010); that abstract was accepted. To my chagrin, instead of the typical complete CAIS-paper, this year someone had decided to allow only what they called “extended abstracts,” which gave one precious little space. Nevertheless, I gave a presentation during the conference, and the “extended abstract” (Smiraglia, “Self-Reflection, Perception, Cognitive Semantics: How Social is Social Tagging?”) is in the proceedings, here: http://www.cais-acsi.ca/proceedings/2010/CAIS055_Smiraglia_Final.pdf.

This was pretty exciting for a couple of reasons. One was that the Globe and Mail got wind of it and kept asking for more text. Unfortunately all I had was the extended abstract, which must not have been enough because I never saw myself quoted. Still, for a moment there I was flirting with the thrill of being reported in the press. As I say often, oh well.

The research itself was exciting enough however. The fictive scanning was there, although once again in small proportions–less than 1% of the total. But more important was the extension of this notion of social classification. It turned out that all of the sites in the study had clusters like those we saw above. In fact, most of the tags were somehow or other associated with the bandwagon effect. There were typically 4 or 5 clusters per site, 2/3 of the tags fell into the clusters, and 1/2 of the tags fell into the two largest clusters. Voila, classification that is social.

I really want to go make a cup of tea but I suppose I should finish with the conclusions, which were:

The taggers collectively are generating a classification with a social basis.

Also, the clusters are not mutually exclusive, demonstrating that a natural classification is not necessarily either hierarchical, or mutually exclusive. But it does remain collectively, potentially, exhaustive.

Warrant becomes a new issue in such a classification, because there is no accountable literary warrant—rather warrant is cultural (as Beghtol predicted).

Those look like some interesting hypotheses for future research to me.

I suppose I should write this up for a journal. But what I really want to do now is look for the same effect on more social social-networking sites.