How to Generate Knowledge Graphs with AI

Knowledge graphs can be very useful for representing the main ideas in a certain field and learning how different concepts relate to one another. They can also play an important role in improving AI LLM workflows giving the model additional context about relations and ontologies which allows it to produce better results (via GraphRAG / HyperRAG implementations).

Another advantage of having a knowledge graph is that it can then be used together with the built-in AI system in InfraNodus to generate insights that are relevant in a specific context you built the graph for. So the knowledge graph becomes sort of a prism through which you can look at any subject.

InfraNodus has a capacity to build a knowledge graph automatically for you. It offers two modes: 

1) Automatic knowledge graph creation for a certain topic (best for overview)

2) Step-by-step knowledge graph creation (best for learning and developing an idea)

Here is how it works:

Generating a Knowledge Graph for any Topic with an LLM

In this mode, InfraNodus will use its built-in LLM workflows to generate a number of statements on the topic that you choose. We use a high temperature and creativity setting so that the responses are diverse and cover as wide range of ideas as possible.

1) To use this mode, go to the https://infranodus.com/apps and then choose Brainstorming / Ideas > AI Generated > Knowledge Graph and type in a topic you want to explore, for instance, "knowledge graphs":

2) After those statements are generated, we use our entity detection system to highlight the most important concepts and entities inside those statements. They will be visualized as the nodes in the graph. The statements where they appear are the relations. The resulting graph will use network science metrics to highlight the concepts with the highest influence (based on betweenness centrality). These will be bigger on the graph. We will also highlight the clusters of concepts that belong together, which we call topical clusters (based on community detection algorithms). These will have have a specific color and be aligned closer to each other.

As a result, you will get a representation that adds all those structural insights and helps you get a good overview of a topic:

You can learn more about the processing algorithm in the article on entity detection.

You can also learn more about reading and interpreting knowledge graphs.

3) As you can see, we can easily see the main topical clusters related to the concept of "Knowledge graphs" are:

- Knowledge mapping

- Decision dynamics

- Narrative insights

- Cognitive complexity

You can choose each of those topics, highlight the part of the graph that relates to it, and see the actual statements, or generate a summary of each topic for you using the AI module.

You can also generate a sumary for all the topics, similar to how Microsoft's GraphRAG works by clicking AI: Summarize Topics button. It will then generate a summary for each topic for you.

4) Zooming in

For instance, while it's clear that knowledge graphs can be used for knowledge mapping and decision dynamics, it can be interesting to see what is meant by the "narrative insights". We can click this cluster and ask the AI to generate a summary of that topic only:

Interestingly, here the summary is telling us how knowledge graphs can be used to generate novel narratives and find new connections across different disciplines. A very interesting use case for knowledge graphs, indeed!

5) Finding content gaps

We can also find content gaps in the graph. These are the topics that are not so well connected. Bridging those gaps can help us generate new ideas and optimize the knowledge graph structure so that all the different topics are better connected. To do that, we can go to the Blind Spots tab in the Analytics panel and ask about the content gaps in this knowledge graph:

We can see that there's a disconnect between "data synthesis" and "analytical correlations", so we can think of a possible connection ourselves and generate a new idea or use the AI to generate a question for us, which we can then use as a prompt. 

By the way, all those workflows are also available via the InfraNodus API, so you can integrate them into your own AI workflows.

Generating a Knowledge Graph Step by Step

You can also generate a knowledge graph step by step using InfraNodus. The difference from the workflow above is that you will actually be generating one statement at a time, verifying it first, editing it if necessary, and then adding it to the graph. 

This approach can be better for learning about a certain subject and having a better control over the knowledge graph that is built. 

To use this mode, go to https://infranodus.com/apps then choose Brainstorming > AI Generated > Step-by-Step ideation:

2) Then type in a prompt, for instance: "How can knowledge graphs be used for developing narratives?" (to continue the previous topic). You will see a representation like this that just contains the node "knowledge graph" and an idea generated in relation to your original prompt, which you can edit and / or add into the graph. 

3) Once you add this idea into the graph, entities will be automatically extracted and you will see a graph:

4) You can then use the advice generated automatically in the AI module or select the parts of the graph (nodes or topics) you're interested in and develop those ideas further. For example, we can select the concepts "nodes", "tool", "narratives" and "understanding" to better connect those ideas (e.g. how can knowledge grpahs and nodes be used as tools for understanding narratives)

We then click the "Response" button in the AI module to generate a statement that would connect those ideas better and we will have a graph that is more coherent and that helps us gain a better understanding of those concepts: