In conjunction with traditional techniques, methodologies based on artificial intelligence can complement qualitative data analysis. These methodologies include text mining based on machine learning, to extract information from textual material. This is shown in a study by three researchers at Fundação Getulio Vargas’ Sao Paulo School of Business Administration (FGV EAESP) – Eduardo Henrique Diniz, João Luiz Becker and Henrique Pontes Gonçalves de Oliveira – in collaboration with Carla Bonato Marcolin of Uberlândia Federal University, published in Qualitative Research in Organizations and Management.

To compare the analysis of texts by humans and machines, the researchers applied automated data analysis techniques to 25 interviews previously analyzed by traditional methods, seeking to highlight the similarities and differences between these means of treatment.

In the context of reusing data generated by other studies and under human moderation, this technique has the potential to support and deepen results previously found, reveal new evidence and reduce human biases. A new methodological model combining traditional research means and big data is proposed as a strategy to enrich such analyses.

The original study from which the data was collected aimed to understand the international development of graduate research in Brazil. The 25 interviews conducted to achieve this goal were analyzed using traditional qualitative research methods, resulting in the development of a model of categories containing the interviewees’ main perspectives and strategies for the international integration of Brazilian research. Through the AI-based analysis carried out later, a new model was generated and compared to the results of the traditional method.

The authors point out that the new analysis was able to identify the same topics revealed by the traditional analysis, confirming two of the four categories previously identified as relevant to the topic of international research integration and distributing the others into new sets equivalent to one or more original topics. The AI model also revealed a new category considered relevant that had not been adequately covered by the researchers in the initial analysis (the importance of the English language) but did not detect one of the original topics (the evolution of the subject over time and its future prospects). 

Based on the comparison between the two models, the researchers developed a new model with contributions from both analyses. The final model kept the initial categories, but with a deeper look at the set of words associated with each one, the addition of a new category and a different hierarchical organization, considering the changes highlighted by AI. In this context, AI-based analysis stood out as a complementary tool for qualitative research. In future research, the authors suggest testing the new methodology on different sets of interviews and improving the technology in order to reduce possible errors. 

See the full article here.