A Semi-Automated Approach to the Content Analysis of Experience Narratives - Modeling Users' Experiences with Interactive Systems - page 127

Fig. 6.6 Visualizing relations between product qualities and product features. Each column
depicts the distribution of experiences over the five product qualities for a given product
feature.

i.e. product qualities. This may lead, for instance, to the identification of the features
that induce desired experiences such as the ones related to self-representation or the
features that are dominated by learnability and long-term usability problems. By
clicking on a cell, the full set of narratives relating to the respective product quality
and product feature appear.

6.4 Validation of the Proposed Approach

Three distinct approaches were tested on a subset of the data in chapter 4. These
were (a) Computing similarity on explicitly defined dimensions using the approach
described in this chapter (Concept Analysis), (b) applying LSA on a restricted list
of terms, and (c) applying LSA on all terms (traditional LSA). Thus, by splitting the
proposed approach in two distinct procedures (a and b), we are able to distinguish
between the impact of i) restricting the list of terms, and ii) defining explicit relations
between concepts and the observed terms.

6.4.1 Preparing the Dataset

These amounted to a total of 347 experience narratives classified under five main
categories: stimulation, learnability, long-term usability, usefulness and social ex-
periences . Figure 6.7 displays the frequency and cumulative percentage of narra-
tives across the number of words contained in each narrative. Large variances in the
number of words across different narratives is likely to impact the computation of
the semantic similarity of narratives with long narratives receiving higher similar-
ity scores to other narratives as the probability of occurrence of words of interest
is greater. This bias may be counterbalanced by weighting frequencies a i,j by the
length of the document j . However, narratives below a certain number of words
are not likely to contain adequate semantic information and may be excluded from