Using Text Network Graph to Generate Better ChatGPT Prompts

ChatGPT is a great tool but its conversational nature means that you have to come up with complex prompts to make it less generic.

Using the text network visualization tool InfraNodus you can use a text graph to steer the conversation in an interesting direction. In this tutorial, we will demonstrate some of the techniques you could apply to your ChatGPT workflow in order to enhance the output that you get from the conversation.

1. Start the GPT Chat App

Go to https://infranodus.com/apps and choose the GPT Chat App

2. Converse with ChatGPT and Visualize the Results

The built-in GPT app in InfraNodus is using both GPT 3.5 and GPT 4 models. Choose the GPT 3.5 model for faster output or GPT 4 model for more precise (but slower) output.

The text you add and the answers you get from GPT model will be visualized as a graph. The words are the nodes and the co-occurrences are the connections. This lets us build a text network graph representation of your conversation and apply advanced network analysis algorithms to detect the main concepts, the most important clusters, the relations and the gaps between them. 

You can then use this graph to improve your conversational flow in the following ways.

Note, the more influential ideas are shown bigger on the graph, while the ideas and concepts that occur in the same context and belong to the same topical cluster will be positioned closer to each other on the graph and have the same color.

If you want to learn more about the algorithm we use, please, read How InfraNodus Works.

3. Develop underrepresented concepts

Use the graph for direct visual feedback on the conversation you're having. As soon as you detect an interesting cluster that you would like to explore further, just click on some nodes on that graph. The ideas will be copied into your chat box and you can send them to ChatGPT so that it elaborates further on these concepts and connects them in an interesting way: 

Once you add them as a prompt, you will force ChatGPT to come up with a response that will be relevant to the conversation but focused on these ideas that you selected, so that, in the end, it will develop this cluster of ideas further. 

This is a great approach you can use not only with ChatGPT but with any text and conversation in general: observing the discourse and trying to bring up the topics that seem relevant but underrepresented to you.

In the end, if you want your discourse to be diverse and ecological in the sense that not one idea overtakes the rest, you want to steer it from a highly centralized discourse to a highly dispersed one:

→

4. Identify and Bridge the Structural Gaps

Once you develop your conversation further, you will notice that it has structural gaps between different topical clusters:

These gaps indicate the parts in your conversation where you could potentially come up with really interesting ideas because they connect the clusters that you already covered but in a completely new way. 

Identify and select these gaps and then think of a question to ChatGPT that would bridge those ideas in a new way. 

Alternatively, use the built-in GPT AI Research Question functionality to have AI generate this question for you. Then feed it to ChatGPT so it has to come up with a response that will bridge these ideas together.

This is a great way of generating an interesting prompt that will steer the conversation in an interesting direction.

5. Exploring the Periphery of Your Conversation

Another interesting approach is to focus on the periphery of your conversation. The nodes and the concepts located at the periphery of the graph are literally on the outer bounds of your discourse.

A good way to interpret this is to think of some tangent topics you touched upon but did not really cover extensively. These ideas may help you think beyond the boundaries of this particular conversation and develop it further in an interesting way, perhaps, connecting it to other interesting topics.

To do that, zoom into the graph and choose the peripheral nodes that seem relevant to you. Then select those nodes and send the prompt to ChatGPT to have it generate a response that will develop these ideas further:

The response will develop these peripheral concepts and topics further, helping you expand your conversation in an interesting direction that seems relevant to you. 

6. Focus on Discourse Connector Points

The next step is to focus on the discourse connector points. These are the concepts inside the graph that have a high influence but not so many connections. Normally, these are the words that connect to a diverse range of topics per the number of connections they have. A good analogy is the social networks: this would be the people that know not so many people but the people they do know are highly important.

Focusing on those nodes helps you identify the parts of the discourse that could connect to important ideas in a highly efficient way, so generating a content that relates to those topics can be highly interesting for opening up the conversation to external ideas. 

You can access it through the Analytics Panel > Gap Insight > Discourse Connector Points. 

You can also just look at the bigger nodes on the graph that are close to the periphery. 

7. Use GPT to Generate Topical Overview of Your Conversation

Finally, you can also use the built-in GPT to generate the names for the main topics that occurred in your conversation. This is very helpful to get an overview and to understand whether you missed out on something important.

Once you get the names for the topics, you can zoom out and look at your conversation from the bird's eye perspective to see if you're missing something important. 

For instance, you may discover a smaller topic that should be developed further. Zoom in, select some nodes, and have ChatGPT develop the idea further. 

...

Try this out on www.infranodus.com