A Semi-Automated Approach to the Content Analysis of Experience Narratives - Modeling Users' Experiences with Interactive Systems - page 118

6.3 A Semi-automated Approach to Content Analysis

In this section we propose a semi-automated approach that aims at addressing the
limitations of Latent-Semantic Analysis in assessing the similarity between self-
reported experiences. The approach is different from Latent-Semantic Analysis in
the following respects.
First, only terms relevant to the phenomena that the researcher is interested in
are employed in the similarity measure. The approach exploits existing domain-
specific knowledge in identifying relevant concepts, but also acknowledges that such
knowledge will always be incomplete; thus the intervention of the researcher in
extracting additional relevant concepts from the data is crucial.
Second, through manual coding the researcher is able to define explicit relations
between concepts, i.e. latent dimensions, and individual measurement items, i.e.
terms of phrases. For instance, an researcher may wish to code when the participant
refers to a relevant person in a social context. The researcher may then create a con-
cept named 'relevant others' that is identified in the text through various individual
terms such as 'friend' , 'brother' , 'colleague' etc. Latent-Semantic Analysis would
not be likely to identify these terms as semantically similar as these would not likely
occur in the same documents. In contrast, through manual coding the researcher is
able to explicitly relate these terms as different manifestations of the same latent
concept.
Third, the approach utilizes visualization techniques in assisting the researcher
in identifying the relevant concepts. For instance, an interacting visualization of
the dissimilarity between narratives enables a systematic comparison of diverse
narratives.

6.3.1 Incorporating Existing Domain-Specific Knowledge

The procedure starts by incorporating existing domain-specific knowledge. In char-
acterizing users' experiences with interactive products, relevant knowledge might
be found in psychometric scales measuring subjectively perceived product qualities
(e.g. Hassenzahl, 2004; Lavie and Tractinsky, 2004; Hornbæk and Law, 2007) and
emotional responses to products (e.g. Desmet and Hekkert, 2007).
Table 6.1 illustrates the five latent concepts that were used in the current study.
The concepts were derived from (Hassenzahl, 2004) and (Lavie and Tractinsky,
2004). Each construct is measured through a number of individual (bi-polar) se-
mantic differential scales (Osgood et al., 1957). Hassenzahl (2004) distinguishes
pragmatic and hedonic quality in interactive products, while Lavie and Tractinsky
(2004) differentiate between classic and expressive aesthetics . Both poles of all in-
dividual scales are used in defining each respective concept, after being stemmed to
their root form (Porter, 1980).