The Semantics of Tagging - Social Semantics: The Search for Meaning on the Web - page 130

5.3.4.3 Tag Suggestions Do Not Cause Tag Stabilization

This experiment provides a first step that leads to a new interpretation of the
accepted theories and models that explain the emergence of power-laws in tagging
systems. Common wisdom in tagging suggested that the power-law was unlikely to
form without tag suggestions. As put by Marlow, Boyd, and others, “a convergent
folksonomy is likely to be generated when tagging is not blind,” blind tagging being
tagging without tag suggestions (2006). The results show that the tags of users
without tag suggestions converge into a power-law distribution. Moreover, a power-
law function fits more closely the behavior of users when the users are not given
tag suggestions than when the users are given tag suggestions. This means that tag
suggestions distort the power-law function that would already naturally occur when
users tag blindly without tag suggestions. These results are not unexpected. After
all, words in natural language naturally follow a power-law , and there exist purely
information-theoretic arguments why this is the case (Mandelbrot 1953).
This helps clarify a number of experimental results from previous experiments
in tagging. First, this result clarifies how the power-law distribution was observed
by Cattuto even before del.icio.us began using tag suggestions via the tag interface
(Cattuto 2006). Second, it also helps explain how the majority of users in Suchanek
et al.'s experiment had a high matching rate, even when in their report-back most
of them said they didn't use or even notice tag suggestions (Suchanek et al. 2008).
Our experiment does have a number of limitations, in particular our experiment
should be extended to deal with more web-pages as well as expert and non-expert
users dealing with different kinds of expert subject matters. In this situation, tag
suggestions may have more of an influence on tagging behavior. Although the
presented results indicate that some of the previous assumptions underlying the
emergence of power-laws do not hold, a power-law distribution alone does not
provide the necessary information needed to determine the role of tag suggestion
on tag behavior. One line of research that seems promising is to understand how
humans categorize in general, which could easily influence how they decide which
tags to use to annotate web-pages. While the large amount of tagging data on the
web made it easy to develop simple mathematical models of human behavior, it
seems that a more detailed understanding of what users are actually doing is needed,
the role of language in the use of the Web by human agents. Therefore, we need to
inspect the collective use of language in tags more thoroughly to get a grasp of what
is occuring with tagging systems as a kind of sense.

5.4 Constructing Tag Correlation Graphs

While earlier we have discovered the kinds of tag frequency distributions that
emerge from the collective tagging actions of individual users, as well as the
dynamics of this process of sense-making, we have come into a key problem.
If the tag stabilization simply reflects the large-scale dynamics of English language