Qualitative Research with Code?

This is a brand new concept I found in research. I have just been exposed by the fact there is this method that shows us how to extract a meaningful cause and effect from just an interview. This research is mostly used in social and humanity faculty but still, this shocks me a lot. It is called qualitative research. Instead of using a questionnaire, it uses and delves into the expected result from deep interviews. How come?

The reason that makes me shocked is how this method is being conducted. The researchers use this method with "coding". Wait, for me that is used to code a lot for machine learning, it is shocking. A code? It turns out the code that is mentioned is a process to categorize something from the theme of the research. They implement this to gain a meaningful result that supports their theories. Amazingly, this method is conducted by the researcher manually. In my opinion, it has to be so boring to do so. Although, there is a way to automate it using software (like NVIVO) and it costs at least 1000 USD. Indeed, it is crazy. So I come up with this idea. We can actually automate it. There are two main things, speech recognition, and categorization, so-called "coding", according to answers from the interviewees.

The first one is speech recognition. The researchers that I met made the transcript of their interviews manually. The first time I heard it, spontaneously said "Arrgh, do we have to?" Indeed, it is important to other researchers in the future if they want to review it. So why just not use speech recognition? Again, it is pricy. My solution is to use transfer learning from a speech-to-text (STT) pre-trained model. One of them is Silero since it can be used in several languages. This one is relatively easy.

The second one is challenging for me. How can we interpret the answer from the interviewees into a categorization and coding system. It is much easier for us if have a chance to set it first. Then, we just have to face the quotations and matrices process. So we have to build a machine learning model. From an interview with each respondent, the machine learning has to quote and pick which answer belongs to which category.

I will elaborate on this through some examples. Unfortunately, I can not do it know since I have to prepare some visualization first haha. That's all for now. Thank you!