Tải bản đầy đủ - 0trang
A. Data Tables, Graphs and Algorithms
Figure 5.3.: Generic workﬂow design of the V-TM framework for QDA
integrating Text Mining: Single steps might be realized by
diﬀerent technologies and under diﬀerent methodological
assumptions to follow either inductive or deductive research
strategies, or a combination of both.
5. V-TM – A Methodological Framework for Social Science
(step 5). Here also certain external data could be taken into account,
e.g. for correlating patterns extracted from a document collection
with text external observations.5
Analogue to the example study, Figure 5.4 presents the schematic
visualization of a speciﬁc V-TM workﬂow chain. It can be perceived as
an instantiation of the generic workﬂow design (Fig. 5.3) integrating
all analysis steps, both inductive and deductive. Additionally, it shows
results as resources produced by one step and utilized as input for
subsequent processes to achieve following analysis goals. In accordance
with the V-Model’s emphasis on testing, each sub-goal needs to be
evaluated separately to ensure the overall quality of results within
the entire workﬂow chain. The principle ‘garbage in, garbage out’
governs any of the applied sub-processes. Serious ﬂaws produced at the
beginning of the process cannot be compensated by later processes. To
assure quality of the entire analysis process, measures and procedures
for quality assessment have to be distinctively deﬁned for each goal.
Deﬁning these evaluation procedures and suﬃcient quality criteria is
a decisive part of any workﬂow design. It guarantees that validity of
a procedure depends on its implementation or parameters only, and
not on variances in quality of its input. Consequently, the example
workﬂow design presented in Fig. 5.4 mentions types of evaluation
procedures and makes suggestions for concrete approaches.
As we are on the second level of the V-TM framework dealing with
high-level research design, the following workﬂow descriptions operate
on classes of algorithms rather than mentioning speciﬁc algorithms.
Input data and outcome of results are given along with substantiated
imperative descriptions of analysis tasks as steps to achieve the analysis
goal. The decision for speciﬁc algorithms and their key parameters
would be part of the subsequent level of TM process speciﬁcations.
Nonetheless, in some cases, I refer to speciﬁc algorithmic approaches
Petring (2015), for example, cross-correlates patterns in news coverage related
to topics of social justice and inequality with the development of the Ginicoeﬃcient in Germany. In a time series analysis, he can show a signiﬁcant
relationship between economic inequality expressed by the Gini-coeﬃcient and
increases of media reports on the topic, peaking after ﬁve months.
5.2. Workﬂow Design