Knowledge graphs can be very powerful for representing relationships in any data. However, once the size of the graph grows, it can be hard to generate interesting insights from it. Network analysis and various metrics from graph theory can solve this problem and this is exactly what InfraNodus is built for.
This can be useful not only for researchers, but also for AI agents, which can get additional structural insights for knowledge graphs, memory stores, and ontologies. Network analysis of knowledge graphs reveals the main hubs, the most influential clusters, and gaps between them. Bridging the gaps helps optimize knowledge and develop new ideas. Knowing the structure helps reveal patterns and provide better context to humans and AI models.
Below is a step-by-step tutorial that will help you (and your agents) learn to use network analysis to read, interpret, and optimize knowledge graphs. It follows the cognitive variability framework: shifting perspective scale and intent.
For this example, we will use an AI-generated ontology of Ray Dalio's economical principles — feel free to open this graph in a new tab to use it as a reference with this tutorial. You can generate your own ontology once you sign up and go to the infranodus.com/apps page for any topic or document. You can also use the InfraNodus MCP server to generate it for you through your favorite LLM tool (Claude, Grok Bot, etc.)
1. Getting an Overview of the Knowledge Graph's Structure
This is the first step that helps you get a general overview of any knowledge, text or ontology, revealing the patterns and the relations inside. InfraNodus builds a graph of relations between the main ideas and reveals the main topical clusters.
Think of this step as if you were building a social network of your ideas: which concepts like to "hang out" together, which live at the periphery, which groups could benefit from connecting to one another?
Steps to replicate: Create your graph or open the existing one above and just look at the overview
Difference from standard AI tools: It's difficult to get this kind of insight using a traditional AI chat interface where the main topics are simply listed without any contextual information: you don't know how they're connected, what the peripheral topics are, where the gaps can be found. Graph representation helps you go beyond generic insight and develop a truly original understanding of any knowledge.
Main Insights from the Graph Structure:
Topical overview shows you the main clusters of ideas and how they connect to one another. This provides a general overview of the structure. You get a high level view of the main ideas inside.
- The main [[entities]] in the graph are grouped by co-occurrence and form clusters ranked by influence. The more influential they are, the bigger they are on the graph. Each cluster has a specific color. Connections between them reveal the most influential patterns.
- Structural analysis panel at the bottom of analytics panel shows how optimal the connectivity is. If the graph is dispersed, bridge the gaps between the clusters. If it's too focused or biased, develop the smaller clusters and go beyond the periphery.
- If you use AI agents, the graph gives them a general reproducible understanding of the whole context in a very condensed form.
How is the knowledge graph built and represented?
InfraNodus will generate a list of statements using [[wikilinks]] to tag the main entities. Each statement is a rich description of the relations between the entities tagged with the relationship type. You can change this syntax by providing additional prompt when the graph is created. Standard text graphs will use lemmas as nodes.
Each relation will be added into the graph where the [[wikilinks]] are the nodes and the edges are the relations between them (described through the statements)
Co-occurrences of the [[entities]] words are represented as the edges that connect the nodes in the graph (the graph is directed).
A special Force-Atlas layout algorithm is applied to group the nodes / words that co-occur together while pushing the most connected hubs away from each others. This generates a readable 3D/2D spatial representation of communities (clusters) that are distinct from one another.
Additional community detection algorithm is applied to group the nodes by clusters. Each cluster gets a unique color and id.
The nodes (entities / lemmas) in the graph are ranked by influence using betweenness centrality — how likely they are to be at the shortest path between two randomly chosen nodes in the network. Betweenness centrality is higher for nodes that bridge different clusters together and measures global influence. The higher is the number of highly influential nodes inside the graph, the bigger is the cluster they form on the graph.
2. Zoom Into the Clusters
Once you have a good overview, you can zoom into the specific clusters to explore each of them in detail. This helps you explore the main topics quickly and get a deeper understanding of the specific topics you're interested in. The graph also shows the specific concepts that belong to the topical cluster you selected.
This ability to zoom into the specific topical cluster helps you read the structure in a non-linear way, zooming into the relevant topics for your research and finding novel connections quicker.
Steps to replicate: Open the graph, then select a specific topical cluster, then click "AI: Summary"
Difference from the other AI tools: Graph helps you zoom into specific reproducibleclusters of ideas that are generated algorithmically, avoiding AI hallucinations. You can then see the source of truth (content related to the cluster) as well as generate granular insights for specific ideas only without polluting the model's context with unrelated content.
Main Insights from Zooming In:
- You get to verify whether the knowledge graph's content covers all the specifics
- You can ask the AI to develop this cluster by adding more ideas
- You can see the content that was used to build the cluster
- You can understand a certain aspect quickly without having to read lengthy AI-generated summaries
- You can steer your model's reasoning towards specific clusters of ideas
3. Find the Gaps
Some of the clusters have gaps between them. Focus on these gaps to find novel connections and generate interesting new ideas and optimize the knowledge structure.
Bridging the gaps between the clusters helps you understand the topic better while also optimizing the knowledge graph to ensure that it doesn't miss important ideas.
Steps to replicate: go to Analytics > Gaps, then click the circle between the clusters and choose AI: Question to generate a research question that bridges the gaps
Difference from the other AI tools: most LLM tools provide the most predictable output (that's how they're programmed), so the ideas they come up with will be very generic. Gap analysis helps you find the ideas that are missing, generate questions so you can stimulate your own imagination, and then use AI to help you bridge these blind spots with new ideas.
Main Insights from Gaps:
- Optimizing the knowledge structure by bridging the gaps
- Better understanding of a topic through interactively generating research questions (can help researchers and LLM agents reason better)
- Finding blind spots in any domain
- Steering the AI models towards more original and less generic ideas
4. Explore the Relations
Switch to the Concepts view to explore relations between the ideas. This helps you see how the different clusters connect and provides a structural understanding of the domain you're analyzing.
Use the detailed graph view to improve your knowledge graph, get a better understanding of relations between the concepts, and optimize the overall structure.
Steps to replicate: go to Analytics > Concepts view, then click any two nodes to explore the relation between them
Difference from the other AI tools: treat this graph as a reasoning map for your AI, edit it manually if necessary, and then use it in your favorite LLM tools
5. Use the Graph as a Reasoning Map in LLM Tools
Once you finished exploring the graph, use it in your favorite LLM agent via the InfraNodus MCP server. The graph structure will be available in your workflows and you'll be able to use it to steer your agents' reasoning. 
Once you inspected and optimized the graph, it will provide clear verifiable reasoning rules to your AI agents. It can also be used to save important procedural knowledge, for LLM memory, rules, and learnings.
Steps to replicate: click the Brain icon next to the graph's name and follow the instructions to add it as a reasoning graph to your LLM
Difference from the other AI tools: instead of writing a plain-text prompt or skill, you have access to a graph (from any client or session) that provides a clear set of principles and rules your reasoning agent should follow in a specific context. You can verify, edit, modify, and change the graph at any time using a visual interface or the InfraNodus MCP server.
6. Save Your Insights to Project Notes
As you are building or exploring the graph, you can save your insights into the Project Notes. These can later be used to build new graphs or as additional system prompt for LLMs that use this graph to augment their reasoning.
Think of the graph as a tree of knowledge. The Project Notes folder is the basket where you keep the best fruits in order to use them to generate new interesting ideas.
Steps to replicate: click the Project Notes folder and write down your thoughts or use the Save button from the AI panel.
Difference from the other AI tools: instead of dealing with a long chat that becomes messy as the context grows, you only select the most interesting stuff and save them to a separate folder, which you can then reuse to generate new ideas
7. Bonus Track: Insight Clusters
Read more about using insight clusters for generating original insight.
Comments
0 comments
Please sign in to leave a comment.