The Semantics of Tagging - Social Semantics: The Search for Meaning on the Web - page 123

Fig. 5.6 Experimental design

Figure 5.6 shows the experimental design. In the 'no tag suggestion' condition
(Condition A), as shown in Fig. 5.6 , a user is presented the 11 websites he needs to
tag without any form of tag suggestions. In the 'tag suggestion' condition (Condition
B), also shown in Fig. 5.6 , a user is presented the 11 websites with seven suggested
tags. While the details of the tag suggestion algorithm applied by del.icio.us is
unknown, for our experiment the suggested tags in Condition B were aggregated
from del.icio.us and the seven suggested tags given by del.icio.us for each of
the 11 websites. For the experiment the seven popular tags were aggregated and
presented to the participants in a manner similar to how tags are suggested to
users of del.icio.us, being shown to the user before they commence their tagging.
Each of the 300 participants was randomly assigned to either the 'tag suggestion'
or 'no tag suggestion' condition. Of these 300 users, 78 did not tag any website
(37 in the 'tag suggestion' condition, 41 in the 'no tag suggestion' condition) and
are therefore excluded from further analysis. The users were randomized over age,
gender, computer, experience with the Web, and their past tagging usage.

5.3.3 Results

In total the 222 participants applied 7,250 tags over all websites in both conditions,
with 3,694 tags applied in the 'tag suggestion' condition and 3,556 in the 'no
tag suggestion' condition. On average every user in the 'tag suggestion' condition
applied 32 . 69 ( S . D . = 9 . 77) tags over all 11 URIs and for the no tag suggestion

conditions 32 80) tags over 11 URIs. Following the same methodology

as done in Sect. 5.2 , we averaged the tag-resource distributions for all 11 web-
pages. This distribution in log-log space is given in Fig. 5.7 . In a log-log scale,
both conditions appear visually to exhibit power-law behavior.

. 61 ( S . D . = 6 .

5.3.3.1 Parameter Estimation via Maximum-Likelihood

Figure 5.8 shows the different parameters for the 'tag suggestion' and 'no tag

suggestion' conditions, as well as the

α

determined via aggregation of tagging
data from del.icio.us for the 11 URIs, using the same maximum likelihood method

α