We all know that good research starts with a question. But how do we know what is a good question to ask?
In this demo, we will show how you can point ChatGPT to any web content using your browser and generate laser-sharp questions that connect the topics in a text in an interesting, novel way. It works especially well for good-quality newspaper articles, scientific research, and YouTube videos.
The secret sauce in this method is the algorithm behind InfraNodus visual text analysis tool. It creates a knowledge graph of a text, identifies the main topical clusters, and finds conceptual gaps between them.
It will then send these gaps along with relevant content to GPT-4 AI (which powers ChatGPT) to generate interesting research questions. You can then generate answers to these questions or even go further and use the questions as an inspiration to chat directly with the web page's content and discover more interesting details.
Here's an explanation of how it works step by step.
Step 0: Activate the InfraNodus Browser Extension
We will be using a simplified version of InfraNodus available as a Chrome browser extension. It has the same capabilities as the standard version, but it is easier for novice users. You can use the advanced graph analysis version as well, but you will then need to use the knowledge graph to generate the insights yourself.
To learn more about activating the browser extension, read the installation instructions on the Nodus Labs website.
Step 1: Find an Article to Analyze
This approach works best for mid- to long-form articles, essays, research papers, and YouTube videos. Basically, any quality web content. However, you could also apply it to index pages of a website, your Jira or GitHub tickets, and much more.
In this demo, we will present some examples ranging from The Financial Times articles to scientific research and YouTube videos, so you see the full scope of what's possible.
Step 2: Generate an Interesting Research Question
For the first example, we will use a scientific review article on the health effects of organic food: Human health implications of organic food and organic agriculture: a comprehensive review (2017).
Once we load the page with the article, we can activate the InfraNodus chrome extension and it will automatically extract the main topics for us. It will then identify the gaps in the content and use the built-in GPT-4 AI to generate an interesting research question. If you're using the advanced InfraNodus version, you need to go to Analytics > Blind Spots > AI Insight Question to generate this question:
This question helps us get into the content of this article in a much more specific way than if we just scan the title, the abstract, and the conclusion. In fact, it does not only tell us what the article is about but also how ideas inside could be developed further.
In this particular case, it's pointing our attention to the fact that organic products might reduce exposure to pesticides.
Now, we can move on to the next step and generate some ideas in relation to this topic.
Step 3: Generate Innovative Ideas
If you put that same text into ChatGPT and ask it to generate an "innovative idea", you'd get a generic answer. Our approach lets you address the gap in the discourse and generate a new idea that would connect the topics that are relevant to the discourse but in a different way.
For instance, for the research article above, GPT-4 yields a verbose response that suggests looking into the effects of organic agriculture on health outcomes. It's an interesting idea, but that is exactly what the paper is about, so it doesn't really advance us further, it just reiterates what was actually said in the paper (and not in the most concise way):
On the other hand, if you click the "Ideate" button in the InfraNodus chrome extension (or "Elaborate" > GPT-4 in the advanced InfraNodus version), you'll see an idea generated based on this question:
Here the idea is much more specific: it suggests how diets leaning towards organic produce may have positive public health implications and reduce medical expenses, which is an interesting practical direction for this research.
By the way, if you check the original article, it does not explicitly propose the link that we discovered above, but it does mention the negative effects of conventional agriculture on public health. The part of reduced medical expenses and, thus, a possibility for promoting organic diets (or subsidizing organic farming) on a government level is what our study came up with!
Step 4: Reiterate Using Q&A Content Chat
In the example above, we pose the question generated along with the topical clusters and top statements from these clusters to the AI to generate an interesting research idea. This means that we "prime" the AI with content that relates to the topics that we're trying to bridge.
InfraNodus proposes another way to get an answer to this question. Instead of taking the statements from the gap clusters, we can also extract the statements from the original text that are most relevant to the question asked. They will not be necessarily related to the topics identified earlier, so it's more likely that they will contain an answer that is already present inside the text itself. It's a great way to see whether our original research question is already answered inside the text and what the answer may be:
In our case, we can see that the text is discussing the effects of pesticides on prenatal development. While it doesn't make the explicit link between public health and medical expenses that we discovered earlier, this answer gives us an important piece of content that we may integrate into our further research. That is, the positive effects of organic farming on public health and reduction of medical expenses using a proven fact that pesticides have a negative effect on prenatal development (and therefore lead to more pregnancies with complications, increasing the public health costs, etc). This helps us make our idea more specific and extract the parts of original content that are already trying to deal with this subject.
Step 5: Extract the Main Topics and Summary
Usually, getting a summary is the first thing people try to do using an AI tool. We think it's best to do it last. Why? Because it might give you a false sense of knowing what the text is about. It is much more productive to start with a question first, generate some ideas, and only then get a summary of what the research paper is about, because you can then link its content to the questions / ideas you generated.
InfraNodus can generate a summary based on the statements found inside the text augmented by knowledge graph and topical structure:
We can see that the main topics are organic agriculture, health effects, pesticide exposure, and omega intake. We can also see from AI-generated summary that it's talking about antibiotic resistance and negative environmental effects of conventional farming. We can add the "omega intake", "antibiotic resistance" and "negative environmental effects of conventional farming" into our new research idea on linking organic farming and public health / reduction of medical expenses.
As you can see, AI-generated summary, in this case, becomes something actionable and immediately useful, as we can integrate it into our research.
Step 6: Reiterate and Reload the Gaps
Once you processed the first structural gap, you can return to the Blind Spots and reload the gap to generate new research questions and topics:
As you can see, this new gap is focusing more on agriculture, antibiotics, and nutrient content — additional topics that we can explore further.
You can then reiterate steps 2 — 5 to extract ideas actionable insight from this content.
Step 7: Add Your Own Content
There is also a possibility to edit the content that's being analyzed or add your own content (using the last tab of the extension or Add a New Graph in the advanced version of InfraNodus). This lets you get this same insight for your own ideas (for instance, the ones that you saved during the steps 1 to 7) and perform advanced visual analysis of those ideas using the InfraNodus knowledge graph.
Interested to try InfraNodus browser extension out? Please, contact us to get your demo copy. You need to be InfraNodus subscriber to be able to access it, so, please, provide your username in the request. Pro and Premium users get a higher AI quota use.