The Semantics of Tagging - Social Semantics: The Search for Meaning on the Web - page 129

Tabl e 5. 2 Imitation rate

URI No. Imitation rate

1 0.22

2 0.35

3 0.29

4 0.35

5 0.20

6 0.34

7 0.31

8 0.42

9 0.50

10 0.48

11 0.43

With prec n ( X , S ) defined as:

i

| T ( X , i ) ∩ S | [ S ( X , i )= S ]

)= ∑

= 1

prec n ( X , S (5.7)

n
i

| T ( X , i ) | [ S ( X , i )= S ]

∑

= 1

The term prec n ( defines the proportion of applied tags that are available in the

single tag suggestion set S . Since the tags S in our experiment are always static,
prec n (

X , S )

is equal to the calculation of the matching rate for the tag suggestion
condition in ( 5.5 ). prec n (

X , S )

defines the proportion of suggested tags that
are available in the tags applied by the user when no tag suggestion is given. This is
similar to the calculation of the matching rate for the 'no tag suggestion' condition.
Therefore we can rewrite the imitation rate as:

NONE , S )

( ) − ( )

mr ConditionA mr ConditionB

ir = (5.8)

1 − mr ( ConditionB )

Tab le 5.2 shows the imitation rates for the different experimental URIs. An
imitation rate of 1 will denote full imitation. The results show that users tend to
select suggested tags when they are available with a chance of 1 out of 3 with a
mean imitation rate of 0.36 (S.D. 0.097).
Combining this insight with our previous work in KL divergence and looking at
Fig. 5.7 , it appears that the 'tag suggestion' condition 'compresses' the distribution
that naturally arises without tag suggestions. This 'compression' of the distribution
that the 'no tag suggestion' generates can be defined as highly frequent tags being
reinforced more and less frequent tags reinforced less or not used at all, leading to
more imitation in the top of the distribution and a 'shorter' long tail. It is because of
this 'compression' caused by tag suggestions that the averaged 'no tag suggestion'
distributions do not significantly fit power-law distributions while the averaged
'tag suggestion' distribution does fit a power-law distribution. Taking a 'scale-free'
power-law as an ideal stable tag distribution, rather counter-intuitively a simple
tag suggestion scheme based on frequency may actually hurt rather than help the
stabilization of tagging as a power-law distribution.