Problem and Purpose
With advances in computing technology, text mining has become an emerging research method in various fields, including bioinformatics (Cohen & Hersh, 2005; Kano et al., 2009; Kostoff, Block, Stump, & Pfeil, 2004; Kostoff, Morse, & Oncu, 2007; Koussounadis, Redfern, & Jones, 2009; Vellay, Latimer, & Paillard, 2009; Winnenburg, Wachter, Plake, Doms, & Schroeder, 2008; Yao, Evans, & Rzhetsky, 2009; Zaremba et al., 2009), business (Consoli, 2009; Miller, 2005; Singh, Hu, & Roehl, 2007; Spangler et al., 2009), engineering (Kostoff, Bedford, del Rio, Cortes, & Karypis, 2004; Kostoff & DeMarco, 2001; Kostoffet al., 2006; Kostoff, Karpouzian, & Malpohl, 2005), and education (Chen, Kinshuk, Wei, Chen, 2008; Huang, Chen, Luo, Chen, & Chuang, 2008; Lin, Hsieh, & Chuang, 2009). The preceding applications have a strong quantitative focus in the sense that the outcome variables can be clearly defined; nonetheless, some researchers have applied text mining into qualitative research projects, and view text mining as a viable qualitative research method (Camillo, Tosi, & Traldi, 2005; Hong, 2009; Janasik, Honkela, & Bruun, 2009).
The purpose of this article is to demonstrate that text mining and qualitative research are epistemologically compatible. First, like many qualitative research approaches, including grounded theory, text mining encourages open-mindedness and discourages preconceptions (Vilkinas, 2008). Second, text mining is similar to content analysis, which is qualitative in essence (Lin et al., 2009). Last, the criteria of good text mining adhere to those in qualitative research in terms of reliability and validity (Krippendorff, 2004).
What is Qualitative?
One may argue that text mining and qualitative methods are vastly different in nature because the former, which employs algorithms for counting words, is inherently a quantitative method. In response to this assertion, Krippendorff (2004) argued that text analysis is indeed qualitative. In his view, reading texts and counting words, regardless of whether it is performed by a human or a computer, does not remove the qualitative nature of the texts. As a matter of fact, today many qualitative researchers employ computer software modules as an aid.
According to Janasik et al. (2009), the seemingly qualitative method of gathering data, such as observation, participation, document analysis, and interviews does not necessarily make a study qualitative. The qualitative attribute of a study resides not in the data collection method, but in the data type and in the method with which the data are analyzed. In their view, in a qualitative study the data should not be converted to numeric values, and mathematical and statistical tools should not be used in the analysis. Rather, the data are processed through systematization, categorization, and interpretation. The first part of the definition (data type as qualitative) is the same as that suggested by Krippendorff (2004), but the second part (the absence of mathematical and statistical tools) is debatable. It is doubtful whether this type of "purity" in methods is an essential feature of qualitative research.
Consider the metaphor of photography. Some film-based photographers complained that digital photographers distort the authenticity of the captured images by digital manipulation, and thus digital photography is computer graphics rather than true photography. However, they overlook the fact that adding filters on the lenses and darkroom manipulation, such as burning and dodging, are also considered manipulation. There is no "purity" in any photographic process. By the same token, purity in the analytical process cannot be a criterion for demarcating quantitative and qualitative approaches. For example, when a quantitative researcher employs exploratory data analysis (EDA) and data visualization (DV) to detect a pattern, there is no "cut-off" value or numeric standard to determine what constitutes a pattern. …