Page 146 - Demo

Chapter 6144and myself ). For almost all of the chapters, I relied on the Google search engine, including Google’s reverse image search. Google’s results are possibly biased and black-boxed (see Rogers, 2019, p. 109). I tried to mitigate this limitation by conducting anonymous searches, cleaning the search history and checking the searches in other search engines such as Duckduckgo. Additionally, the gathering of data using Google and also Twitter API (special thanks to Prof. dr. Art Dewulf for enabling this) is ‘sensitive to keywords’ (see Rogers, 2019, p. 37) and therefore has the risk of not being exhaustive, when relevant keywords are not known to the researchers. Limitations that come with the reliance on automated visual categorisation exist in Chapter 2, where we used software (ClarifAI) that labels visual content by categorising it as humans, landscapes, icons, etc. This labelling has limitations; for example, the software does not consider any cultural context when labelling content. Additionally, some visual content may be ambiguous, in terms of labelling. Whereas in manual labelling (often termed coding), ambiguity can be discussed with others until an agreement is reached, the software simply provides its black-boxed verdict.6.3.2 Limitations to working with interpretive methodsIn each of the chapters, I have been working with interpretive methods. The use of interpretive methods comes with limitations: The construction of coalitions can be done in different ways, and the coding of sentiments, narratives and storylines can also be conducted in different ways. In addition, in Chapter 3, the grouping of visuals based on type and content was mainly interpretive and manual and can be done in different ways. To mitigate these limitations of interpretive research, I checked the textual and visual coding with the team of supervisors and a team of students working on similar topics. We also did this for the type of visuals and included a visualisation expert from the Netherlands Environmental Assessment Agency (Filip de Bois), to enable rigorous and relevant coding.Another difficulty was that there are also visualisations that contain text integrated into them, and I needed to decide how to approach this text: as text belonging to the textual analysis, as an element with only visual qualities (thereby ignoring the readable message) or as an element that has both visual and verbal qualities. I opted for the third option. Selecting any of the other options, obviously, would lead to a different analysis.Efrat.indd 144 19-09-2023 09:47