A Semi-Automated Approach to the Content Analysis of Experience Narratives - Modeling Users' Experiences with Interactive Systems - page 129

50

40

30

20

10

0

10 30 50 70 90 110 130 150 170 190 210 230 250

20 40 60 80 100 120 140 160 180 200 220 240 more

Number of words per narrative

Fig. 6.7 Histogram on the length (number of words) of experience narratives. 5% of the
narratives, i.e. the ones containing less than 24 words, were excluded from further analysis.

terms or phrases in the narratives. Five concepts were derived from existing domain-
specific knowledge (see table 6.1) while the remaining 21 concepts were derived
from the data through the qualitative coding procedure discussed in this chapter
(see table 6.2.
The dissimilarity between narratives was then computed using the 26 concepts
(equation 6.7), resulting in a 329x329 distance matrix.

6.4.3 Latent-Semantic Analysis on Restricted Terms

In the second approach, the explicit relations between concepts and terms were dis-
carded. Instead, Latent-Semantic Analysis was applied using the restricted list of
terms (539) that were identified by the researcher. Singular Value Decomposition
was applied to the 539x329 matrix to extract the 26 most dominant latent dimen-
sions. The optimal dimensionality in LSA is an ongoing research question, with
some suggesting a dimensionality between 100 and 300 (Landauer and Dumais,
1997), while others suggest that most variance can be captured in the first 10 di-
mensions (Kontostathis, 2007). We applied a shared dimensionality of 26 in all three
approaches to minimize any effects induced by differences in dimensionality when
comparing the three approaches.