A Semi-Automated Approach to the Content Analysis of Experience Narratives - Modeling Users' Experiences with Interactive Systems - page 125

6.3.4 Hierarchical Clustering

Once the similarity between all pairs of narratives has been computed, the distance
matrix can be submitted to a clustering algorithm and the emerging clusters can
be characterized by the three most dominant terms or concepts that appear in the
narratives that are grouped together.
For instance, table 6.3 presents the three most dominant concepts for each of the
first nine clusters that resulted from Ward's hierarchical clustering with a minimum
variance criterium on the distance matrix of the 329 narratives. One may note that
cluster 1 is dominated by experience narratives relating to Stimulation, clusters 4
and 8 mostly by narratives relating to learnability, clusters 3, 7 and 9 by narratives
relating mostly to long-term usability, clusters 2, 5 by narratives relating to useful-
ness and narratives in cluster 6 relate to social experiences. This information may
assist the researcher in further grouping the resulting clusters.

Ta b l e 6 . 3 Three most dominant concepts in each cluster

Cluster No. terms.

Cluster 1 Stimulation, Novelty, Aesthetics in Interaction

Cluster 2 Missing functionality, Useful, HQ-Stimulation

Cluster 3 Long-term usability, Efficiency, HQ-Stimulation

Cluster 4 Learnability, Simplicity, HQ-Stimulation

Cluster 5 Useful, Fast access to info, HQ-Stimulation

Cluster 6 Relevant Others, Self-expression, HQ-Stimulation

Cluster 7 Long-term usability, Fits context, Adaptable

Cluster 8 Understandability, HQ-Stimulation, Pragmatic Quality

Cluster 9 Long-term usability, Effectiveness, simplicity

6.3.5 Visualizing Insights

Once narratives have been classified into categories that represent similar mean-
ings, a number of generalized insights may be gained through the visualizations and
exploration of the interrelations of meta-data such as the day in which a reported ex-
perience occurred (out of the 28 days of the study), and a self-reported satisfaction
value (Osgood et al., 1957).
For instance, figure 6.5 illustrates a visualization of frequency and average impact
of four types of experiences over the four weeks of the study. The Y axis corresponds
to the number of experiences being reported for each respective experience type
per week, while the average impact of experiences (as reported by the participants)
for a given type are visualized through the size of each circle. Four overall types
of experiences are distinguished: experiences relating to stimulation, learnability,
long-term usability and usefulness . A 4-point (0-3) impact index of each reported
experience is derived from the self-reported satisfaction elicited through a 7-point