One of the big advantages of AI tools is the ability to have a conversation with any content. It can be especially beneficial for long-form content like YouTube videos, which sometimes take hours to watch. Using the new AI-enhanced visual summarization function in InfraNodus, you can generate a visual summary of YouTube videos for a certain topic as a knowledge graph and use a built-in AI chat function to ask any question related to the content. It can be used for specific videos, YouTube search results for a certain query, whole YouTube channels, and playlists.

Summarize YouTube Videos with a Knowledge Graph

For instance, imagine that you’d like to know what’s currently happening on the topic of AI. There are plenty of videos on YouTube on that topic, but watching them all can be cumbersome. With InfraNodus, you can import the top 20 (50, 100) search results from YouTube for this topic, analyze the content, and represent it as a knowledge graph (in under 2 minutes). You will then generate a general overview of the main ideas and be able to use the built-in AI chatbot to ask the questions you’re interested in.

In our particular example, after we imported the top 20 YouTube videos for the topic of AI, we can see that the most prominent concepts are “ai”, “data” and “model” but on a higher level the main topics are:

• AI Development
• Data Generation
• Tech Industry
• Human Interaction

This is a really good way to find out what’s happening in this discourse at the moment. We can see that it’s quite technical, but also there’s a big topic on how AI can better engage with humans. Using the generated knowledge graph, you can also see various metrics derived from text network analysis: which concepts have more relevance (they’re shown bigger on the graph), which form topical clusters (they’re closer to each other and have the same color), and — more importantly — what are the gaps between those clusters.

For instance, in our case, InfraNodus has identified a gap between the two topics:

• Tech Industry
• Human Interaction

then it proposes a question (the right Analytics panel) that is proposing to make the AI more humane.

It becomes very easy to provide good-quality questions and summaries because we simply need to traverse the most relevant concepts and topics to ensure that AI-generated summarization is pertinent to the content. It also becomes easier to generate good-quality questions using AI, because we can simply target the gaps we identified and therefore ask questions that bridge gaps between ideas.

This approach can also be very useful for YouTube content creators: suddenly, you can identify the most popular topics and find the content niches that are not yet filled.

AI Chat with YouTube Videos

You can then start asking questions — either using the AI-generated ones or your own — using the built-in AI chatbot. For instance, “What is the hot topic in AI at the moment?”

Once you ask this question, InfraNodus will use the underlying RAG system to generate the most relevant parts of content related to your query, as well as the most relevant underlying knowledge graph structure. This is an important advantage it has over other systems because it allows you to supply contextual information to AI models so that they can generate a more precise and pertinent response for you.

For instance, if you ask an AI model the same question, it will just use the RAG to generate the most relevant statements and generate a response from them. InfraNodus, however, will also make sure — in addition to that — that the response is closer semantically to the data you’re working with because it has this underlying graph structure, so you can get more precise responses that relate directly to this content.

For instance, we can see here in our example that a lot of videos are talking about the intersection of AI with personal assistant technology as well as alignment and transparency issues. You can then the AI chat to elaborate further on this answer and provide more practical information on how to evolve these ideas further.

Finding the Relevant Parts of the Videos

Another exciting feature is the ability to select the content you like and jump into it to explore it in more detail. For instance, what would happen if we clicked on some parts of the content that are most interesting to us, such as “human” and “interaction”? InfraNodus will then retrieve the videos and the exact timestamped moments, which talk about those topics, so you can click the link and jump to the specific part of the specific video that talks about those ideas:

Moreover, you can also remove the most “obvious” nodes from the graph to see what’s hiding underneath. This is particularly useful if you want to explore latent topics in YouTube video content and to better understand what’s hiding underneath the surface.

Feel free to try this approach on www.infranodus.com