Supervised machine learning approaches do not have ground truth values. This allows explorative data analysis without preconceived conceptualisations. In this manner, unsupervised machine learning is similar to a grounded theory process, where data-driven classification of the data is done to codes, themes, categories or other relevant larger concepts.viittaa GT paperiin For example, Pantti et al. (2019) used unsupervised machine learning to explore how racism is discussed in the Finnish media environment. Their analysis showed that racism is often discussed in the context of culture and movies in Finnish news media. This was a surprising outcome, and the result was extensively discussed among the co-authors during the data analysis stage. To ensure that the result was not erroneous, various examples were manually inspected and confirmed that they were correctly interpreted.
This case also illustrates the challenges of unsupervised machine learning for data analysis. Algorithms will always provide a solution to the problems. Therefore, the importance of validation is stressed in the literature (Grimmer and Stewart, 2013). Validation is supposed to ensure that the solution represents data. However, the solution requires extensive reflection and interpretation, which brings more subjectivity to the analysis.viittaa juho Therefore, it is keen for various human biases (like many other research methods).