The Semantics of Tagging - Social Semantics: The Search for Meaning on the Web - page 145

5.7 Conclusions

To conclude, user-generated collaborative tags can serve as a digital neo-Fregean
sense, a kind of sense more limited than the vaster notion of sense from Frege,
but a notion with the distinct advantage of being computational and so tractable by
machines. So we are interested in proving the existence of a trace ,a digital encoding
of user behavior for a resource . Using KL divergence, we can show that tagging
distributions per resource do indeed stabilize the scale-free power law distribution,
so that the 'tag cloud' of a resource after a certain point stabilizes into what is
widely-accepted in a particular community to be a good description of the resource.
Furthermore, this behavior of stabilization is a function of time and number of users,
and does not simply reflect an artifact of the tag suggestion mechanism. Tagging can
indeed be the foundation for a sense-based semantics on the Web.
Also, it seems tagging produces a richer notion of sense than search terms. This
can probably be explained by the fact the del.icio.us users have more expertise and
interest in complexity-related topics than general web searchers. Furthermore, they
are probably more careful in selecting resources to tag and in selecting labels for
them that would be useful to other users as well (general Web searchers are known
to be 'lazy' in typing queries). As a caveat, we note that this target domain (i.e.
complexity-related disciplines) is scientific and very specialized. If the target would
be more general (for example, if we selected a set of terms related to pop-culture),
the comparison might lead to different results. Also, people who sign up to use a
collaborative tagging system are implicitly more willing to share their knowledge
and expertise with a community of other users. By contrast, Web search is implicitly
a private activity, where tracing an user's actual identity may not only be undesirable
to the user, but also the user may not even be aware their activity is being tracked so
their keywords can then be used by search engines or other programs to change the
results for other users.
The question remains: while one can operationalize some notion of Fregean
sense-based semantics on the Web in the form of collaborative tags, is this enough?
After all, many URIs are not tagged at all! Superficially, the preliminary results from
search engine keyword analysis seem to show that keywords are a much sparser
source of sense than tags. However, these results only were shown on a tiny group
of keywords gathered from a search engine on a particular topic. To think more
broadly, perhaps all associated keywords with a particular resource could serve as
a better sense-based semantics for a URI. This may include not only the keywords
from tags explicitly given to that URI and from keywords used to reach a URI,
but also from the terms accessible from the Web representations hosted at the URI,
ranging from Semantic Web documents to hypertext web-pages. It is to this more
comprehensive notion of computational sense that we turn next.