5.1. AI chat based analysis
The AI Chat tab in the main window allows you to discuss certain aspects of your analysis with the AI. QualCoder provides four different types of chats:
Note that the AI must be fully loaded and ready for new queries ("AI: ready" in the status bar) before you can start a chat.
On the left side of the Ai Chat tab, you will see a list of all previous chats in your project. Double-click on a chat (or press F2) to edit the title for better organization.
- Gives you a quick overview about a certain topic, concept or phenomenon in your empirical corpus without coding your data first.
- Use this as a first exploration and to familiarize yourself with the data.
- Is based on the same search technology that also powers the AI Assisted Coding.
In the AI Chat tab, click on "New > New topic analysis chat", and the following window will appear:
-
Enter a topic that you want to explore and (optionally) a description. Note that you don’t have to enter exact search keywords here. You can explore much more abstract concepts as well (e.g., "changing gender roles").
-
Select an analytic prompt:
- Topic summary will give you a brief summary of the results, usually in bullet points.
- Analyze unexpected will try to extract new, surprising, and unexpected insights from your data by using the vast common-sense knowledge embedded in Large Language Models as a backdrop for comparison.
- Analyze differences will specifically look for differences between the cases, documents, etc. in the data.
(As an advanced option, you can also define your own analytic prompts by clicking on the “Edit” button, which opens the AI Prompt Library.)
-
As a third step, you can add filters (documents, cases, attributes) to limit the data included in the analysis.
After you click OK, the initial search for related data in the AI's local memory is performed. After a while, the analysis will start to appear on the screen.
The AI will add references to the original sources as hyperlinks (blue, underlined). Move your mouse over the reference to see the underlying data fragment. Clicking on the hyperlink will open the fragment in the text coding workspace so you can see it in context.
Remember that this is a chat: you can ask follow-up questions or discuss the interpretation with the AI. However, there is currently no way to add a new topic or new data to the conversation. If you want to go deeper into a subtopic or explore another aspect, start a new chat.
- Allows you to analyze selected passages from empirical documents with the AI.
- Provides support or additional ideas during the detailed text interpretation phase.
This analysis will first switch to the text coding workspace (Menu: Coding > Code text). There you can open a document, select the text you want to analyze, right click on it and choose "AI Text Analysis" from the context menu (will be greyed out if the AI is not yet fully loaded and ready). A submenu will open, allowing you to choose between different types of text analysis:
- Interactive brainstorming with AI: a question-driven brainstorming session about your data
- Paraphrase and summarize: a condensed paraphrase, no further interpretation
- Themes generation: extraction of a list of themes (based on Friese, S. (2024): Prompting for Themes. https://community.qeludra.com/posts/prompts-for-qualitative-research-prompting-for-themes)
- Analyze the Unexpected: Look for new, surprising, and unexpected insights from your data
- Reconstructive SRP: comprehensive analysis based on the documentary method (Source: Lieder, F. R. & Schäffer, B. (2024). Reconstructive Social Research Prompting (RSRP). Distributed Interpretation between AI and Researchers in Qualitative Research. https://doi.org/10.31235/osf.io/d6e9m)
Note that it is not advisable to select an entire document or a very long section of text for analysis. This will lead to superficial results and can become expensive over time as OpenAI charges you based on the amount of text sent with each query.
If you want to define your own types of text analysis, you can do so by adding them to the AI Prompt Library (advanced option).
- This will analyze all the data collected under a particular code or category.
- Use this in later stages of your research to discuss and consolidate your findings.
- It is similar to the "topic analysis" described above, but works on your coded data, not a free topic search.
From the AI Chat tab, click "New > New Code Analysis Chat". In the following window, select a code or a category you want to analyze, and choose between several analytic prompts:
- Code summary will give you an overview of the data associated with the selected code
- Analyze unexpected will look for the most interesting, unexpected insights in you coded data
- Analyze differences will examine whether there are significant differences between your empirical cases/documents with regard to the selected code.
Note that a code analysis will send all the data collected under that code or category to the AI, up to the maximum input length set for the AI model, which is about 100 pages of text in the case of GPT-4 turbo. At the time of this writing (Nov 2024), sending a 100-page text to GPT-4 turbo for analysis would cost you about 60 US cents. Your actual requests are likely to be much smaller and cheaper. However, keep the cost in mind when doing extensive code analysis with the AI.
- Ask the AI anything that is not directly related to your data – basically a build in ChatGPT.
Use this for testing and for brainstorming with the AI about theories, methodology, or what to cook for the next dinner with the research team. Note, however, that this chat has no access to your data and relies only on the internal knowledge of the language model.
