A Semi-Automated Approach to the Content Analysis of Experience Narratives - Modeling Users' Experiences with Interactive Systems - page 133

of relations between concepts and individual terms, and not in the restriction of the
list of terms to the ones that concern particular phenomena of interest.
At the same time, the proposed approach displays a number of advantages over
traditional manual coding procedures as concepts already defined through the anal-
ysis of a subset of narratives, are indexed and used to characterize all narratives.
This has two main implications. First, in the case of an established body of knowl-
edge in the domain, the researcher can start from an initial classification. Second,
the researcher is assisted by the iterative process of coding and visualization where
the impact of newly defined concepts is directly visible on the full set of data.
This, however, also introduces some complications for the coding process. When
defining a concept through the interpretation of a certain term or phrase, the re-
searcher is also forced to assess the generalizability of the relation between term and
concept under different contexts. In principle, the researcher needs to avoid coding
lengthy phrases as the likelihood of occurrence of the same phrase in other narra-
tives decreases as its length increases. On the other hand, a narrow interpretation of
a single term in a given context entails the risk of false attribution of meaning of
the term in other contexts. Thus the researcher needs to find a balance between, on
the one hand, effective characterization of narratives, and on the other hand, omitted
risks of false interpretation in other contexts. We argue that this process minimizes
the risks of over-interpretation of qualitative data as the researcher is forced towards
a systematic comparison of the use of a given term or phrase across different con-
texts.
While the proposed approach argues for the aggregation of experiences into gen-
eralized knowledge, it does not intend to downplay the role of idiosyncratic insights
in the design process. Instead, it argues for an interplay between generalized knowl-
edge and idiosyncratic insights. The latter ones may provide a rich understanding
on how the product is experienced in a given physical and social context. They may
lead to an understanding of users' needs and generate ideas for design. But, de-
sign needs prioritization. Generalized knowledge may point at the important issues.
For instance, how frequent is a given experience? What experiences dominate over
prolonged use? Next, not all experiences are equally interesting to the designers.
An information-theoretic approach would suggest that interestingness relates to the
rarity of an experience. Thus, by visualizing the interrelation between experience
narratives (see fig. 6.2) we enable the researcher to identify experiences that bear
limited similarity to others and are thus more likely to contain new information for
design.
The proposed approach is only a first step towards more automated approaches
for visualizing and interacting with qualitative data. A number of limitations and
future directions may be noted. First, coding is now restricted to matching terms
or phrases whereas more advanced coding procedures might also include logical
operations. For instance, the researcher might desire to identify a concept when
two terms appear (at any place) within the narrative or when a term appears while
another term is absent (e.g. the term beautiful is likely to relate to the concept of
visual aesthetics only when terms like simple and interactive are absent). Similarly,
axial coding is now restricted to establishing a hierarchical structure between the