InfraNodus can visualize any text as an AI-powered interactive knowledge graph that can be used for brainstorming.
The graph can help you see the main concepts and topics within, reveal the relations and content gaps between ideas, and use the built-in AI to generate interesting research questions and facts that can help you develop the discourse further.
Here is a recommended workflow for brainstorming using the InfraNodus graph.
Step 1: Go to the Brainstorming Mode > Step by Step AI Brainstorming
Go to the https://infranodus.com/apps page and click the Brainstorming tab at the top.
There are two main AI brainstorming modes in InfraNodus.
Then type in your topic of interest (e.g. "machine learning" or "heart rate variability") and then select the ideation mode you prefer:
1. AI-generated knowledge graph or
2. Live Step-by-Step AI Graph Brainstorming
The difference between these two modes is that the AI-generated knowledge graph will produce 10 or more statements on the topic you choose for you and pre-populate the knowledge graph. This can be interesting if you want to start with a general overview and then start adding your own ideas:
The live step-by-step ideation mode will take you through the knowledge graph generation step by step, which can be interesting if you want to have a better control of the ideation process and edit every idea before it is added into the graph:
Note, that the default processing mode will detect entities in your AI-generated text, which results in a sparser, more readable graph (so, "machine learning" will be shown as one node "machine_learning" on the graph).
You can also choose to visualize separate concepts (word lemmas), which will provide more raw context to your AI model but it makes the graph denser and harder to read (so "machine learning" will consist of two nodes: "machine" and "learning").
Step 2: Add the First Idea to the Graph
InfraNodus will visualize the idea you added in the knowledge graph. If InfraNodus detected the phrase you added as an entity, it will visualize this entity on the graph and open the AI module, which will generate ideas related to this entity.
If you like the idea, you can edit it, and add it to the graph. Then InfraNodus will extract entities from that idea and add them to the graph:
Suppose you add this idea to the graph, then you will see the concepts inside visualized as the nodes, co-occurrences will be visualized as the connections. The bigger nodes in the graph are the more influential concepts. The ideas that form topics will have the same color and will be closer to each other. You can use this visualization to start exploring the topic and understanding the important ontological relations within.
Note, that you can also add your ideas manually using the Statements Editor on the right. In this case, you can use the AI for inspiration but formulate the ideas yourself. For instance, you will see the node "machine learning" on the graph and can then think of what else it should be connected to and why. This may be more interesting for learning, but it will take longer and is more difficult.
Step 3: Explore the Graph
Here's what the graph will look like after you add a few ideas (either generated by the AI or your own) into InfraNodus. Note that both the concepts and the main topics are shown in the graph and in the analytics panel on the right.
The words are the nodes and their co-occurrences are the connections between them. Once we build a graph in this way, we can use powerful algorithms from graph theory and network analysis to detect the most influential concepts , the topical groups, and the content gaps between them.
💎 Pro Tip: How to Read a Knowledge Graph:
Here's a brief explanation of how you can read the graph. Learn more in How to Read and Interpret Text Network Graphs article
a. The Most Influential Concepts
The main concepts are shown bigger on the graph. The measure we use to rank them is based on betweenness centrality measure that detects the nodes that connect distinct groups of nodes together. In our case, these are the concepts learn, learning and models
b. Topical Groups
The topical groups are the nodes that are placed closer to each other on the graph and have a specific color. We use the community detection algorithm used in network science to detect those clusters of nodes in combination with Force Atlas layout algorithm to visualize their location.
Both the most Influential Concepts and the Topical Groups are also shown in the Analytics Panel on the right.
The advantage of this approach to traditional content generator tools like ChatGPT or Jarvis is that InfraNodus takes the context of your discourse into account (using the underlying portable GraphRAG system) and extracts the most important elements from it first before asking the AI system to generate a new text.
We can develop it further by generating more facts and adding more content.
Step 4: Using the AI Insight Helper to Generate More Ideas
One of the biggest advantages of InfraNodus is that you can steer the built-in AI to think about certain ideas or topics by selecting them on the graph. The knowledge graph becomes a steering device for thinking, which you can use to explore ideas.
You can select specific concepts or topics in the graph and ask the AI to generate ideas that relate to them.
For instance, in our example after adding the first idea, we have three topics. We can select the two topics and use the AI module's recommendation to bridge the gap between them:
- Algorithmic Predictions
- Learning Adaptations
The default AI mode for idea generation is the "Response" mode. When we select this mode, InfraNodus will generate a statement that links these two topics with a statement:
Click "Save to Graph" if you like the idea and want it saved into the graph. When you click this button, it will be saved with a "ai ideas" tag, so you can later separate it from your own ideas (marked with "ideas" tag). Click "More Responses" to generate more responses.
For instance, in this case, InfraNodus generated an idea that's talking about the quality of AI technology where it is able to adapt to specific scenarios, bridging the gap between algorithmic predictions and learning adaptation. This means that more personalized solutions can be found on the basis of AI that go beyond hardcoded algorithms.
We can edit this AI-generated idea and add it into the graph:
💎 Pro Tip: Choosing the AI Insight Mode
You can also choose other AI modes.
For instance, if you prefer to generate a research question and to then think of an answer yourself (or feed it back to AI as a prompt), you can use the "Question" mode.
- Response — general mode that works similarly to ChatGPT where it generates content related to the topics you selected
- Question — generates the research questions using GPT-4 AI related to the concepts that you selected;
- Idea — generates innovative ideas using GPT-4 AI
- Assertion — (available in the advanced mode) generates interesting facts using GPT-4 AI that relate to the concepts you selected
- Challenge — (available in the advanced mode) will challenge the selected idea
Step 5: Bridge the Content Gaps
One important advantage of using network representation of discourse is the ability to detect content gaps in the discourse. These are based on the structural gaps that appear between the topical groups that could be better connected. Bridging those gaps can help us generate more ideas.
In order to do that, we can go to the Analytics > Gaps menu and ask the AI to generate a research question that would bridge the gap between "Data Dynamics" and "Tech Evolution":
The question is asking us how we can improve accuracy of predictions by feeding the model with future scenarios. This could be an interesting for using machine learning algorithms for stock price prediction based on the discourses about the future.
Think of the answer yourself (we can't let AI do everything for us :) and add it manually; or click "Elaborate" and send this question back to the AI to generate a Response that would bridge this gap. Make sure to untick the "focus on this context" option to push the model beyond the boundaries of this graph:
After we add an idea generated by InfraNodus, our graph becomes more diverse, linking the topic of evolution and data dynamics with an answer that is talking about algorithms and clustering. Making those technical topics more important:
Step 6: Explore the Periphery
Another interesting thing you can do when you brainstorm with a knowledge graph is that you can focus on a peripheral topic. You can detect it directly from the graph: the topics that are further away from the center are less integrated in the existing discourse and have the potential to develop it in an interesting direction.
For instance, the "Regression Clustering" topical cluster is an interesting one, let's zoom in on it and click some of the concepts inside. Then use the AI module to generate some innovative ideas in relation to it:
Here it's proposing a novel idea that leverages a hybrid machine learning model to dynamically shift between supervised and unsupervised learning methods — something that could be implemented in healthcare or climate forecast scenarios. For instance, combining regression-based models (that attempt to predict an outcome of a treatment) with clustering (which categorizes symptoms and treatments) and then deciding on the best possible route of action.
Another interesting way to develop peripheral ideas is to focus on Conceptual Gateways: the nodes that have high influence per the number of the connections. As influence in this network is measured based on how well a concept connect various topics together, a highly influential node with fewer connection will be an important but accessible topical broker, making the concept a perfect candidate for developing the discourse further and connecting it to other fields.
For example, in our case after adding the idea above we identify the following conceptual gateways that we can use to develop our understanding of this topic further:
Step 7: Uncover Underlying Ideas
Once we add more content in the graph, we can also slice off the top layer of most influential concepts to see what's hiding underneath. We can do this deleting the nodes from the graph (slicing off the top layer of concepts) — this can be a very powerful way to access underlying ideas that are not visible on the surface:
After you remove these nodes from the graph, your AI module will focus on the underlying concepts that become more prominent. So you can develop the peripheral ideas better by selecting them or the topics that they belong to and exploring them in more detail. For instance, how machine learning can be applied to improve agricultural sustainability — a cluster we wouldn't have seen behind the more prominent ideas before:
Step 8: Analyze the Graph and Reiterate
Once we are done adding new ideas, we can use the basic Graph Exploration Workflow to reinforce our understanding of this topic and find out more ways to develop it further.
Comments
0 comments
Please sign in to leave a comment.