Automating the QDA process - unleashing AI
Oct 18, 2023, Maniish KarveA couple of weeks ago, I wrote about "Generative AI and Qualitative Data Analysis: New Frontiers?" . In this blog, I elaborated a little bit on how you can experiment with Generative AI yourself, some insights from our conversations with researchers across the globe, and finally, about the need for a co-pilot in this journey.
I am going to close out this series on Generative AI and QDA (do I hear a collective sigh of relief??) by talking about different aspects of automating the qualitative research process in general, and the QDA process in particular.
So let’s dive right in. I will be talking about what directions one can take, and how our experiments are panning out and becoming features on our platform.
Revisiting where Gen AI can make a difference in QDA
Hark back to an earlier blog I had written about Top 3 advances in AI that matter. We had looked at the following visualization of AI and its possible influences in the process.
Let’s focus on the elements 8 – 10. Lets see what is out there.
Automated transcription
AI-powered transcription can quickly and accurately convert audio or video recordings, interviews, into digital text. This literally happens in minutes. Of course different models offer different levels of accuracy for different languages. The good part is that there are enough and more people solving the accuracy bit that should give us confidence to start this journey.
On flowres.io, you can simply upload your own video or audio file, get an auto-generated transcript and then edit it yourselves or through a proofreader. We can vouch for higher accuracy in 40+ languages with custom trained models. And we offer a proofreading service to bump up accuracy for any over 200+ languages. Here is an example of an auto-generated transcript.
Transcript Dashboard
And largely, this is the approach that is common across other platforms as well. We tested a few, and the results vary by language as expected and by accents. All of them provide an editor to fix the transcript yourselves. Some examples for you to try out are Rev, Otter, which seem to be very popular.
The challenge is in translating though, and this is where the flowres algorithm shines – it automatically translates the audio to English during transcription.
Coding text data (and extracting themes)
This is where, as discussed extensively online, a lot of work is currently being done. Thematic coding seems to be the approach everyone seems to be taking (MaxQDA, Envivo etc.), either using proprietary NLP with learning models, or accessing Generative AI frameworks.
Thematic models, in our considered opinion, offer a great deal to the researcher. However the amount of work required to transition the themes to appropriate buckets for reporting seems to be extensive. Of course, some established players such as MaxQDA allow you to import analysis headers, but this has fairly limited functionality.
At flowres.io, our approach has been slightly different. We are working on leveraging generative AI to automatically convert the transcript into a grid if provided with the analysis headers required, or codes if you will. Apart from the evidentiary value of verbatim tagging, the possibility of getting a tagged response with the ability to summarize, or even vice versa with a verbatim citation is important when you want to scale the science part of QDA (it is often said that an experienced researcher is also well versed in the art of extracting insights for reporting)
Summarization and second level analysis
Enough material is out there about how this is being revolutionized with the advances in AI.
Fact is, AI-powered summarization, both extraction based and abstraction based, are on a path of continuous improvement. As AI models become more sophisticated and better at understanding context and human sentiment, their role in summarization and second level analysis will expand for sure.
At flowres, we are looking to help researchers taking this journey at 2 levels.
One, at the level of raw data itself. Here a researcher can engage in a conversation with the transcript or transcripts, and try to get to insights by doing what they do best, ask questions.
The other, working with grids to generate insights with relevant citations that are reporting ready. For example, evaluating all the answers to an emotion wheel and generating frequency distribution and extracting key drivers as well.
So where does the automation come in, you ask?
Well your starting point is the platform itself. Head on over to flowres.io, and try out some of these tools yourself. You can get access through a 30 second sign-up, and then create a project, upload your media and get cracking.
flowres.io
If you want to experience the CA tool again, head on over here.
Experience Zone
If you don't see what you want, let me know- I will demonstrate 'how to' and get you access.
In Conclusion
As I said in a previous post, navigating through this high-speed Generative AI autobahn can be challenging. Not just in terms of keeping track, but for identifying and testing the most relevant developments for your specific needs.
That is where a capable co-pilot can add tremendous value to your endeavors. If you already have taken steps to get on the highway, kudos, wish you success in your endeavor. If you are in DIY mode, or if you would like to know what is the most impactful way for your business to leverage these fast paced developments for QDA, happy to discuss our learnings in detail and help.
