Timecodes: 0:00 Introduction 0:48 Content import: Google search results for AI automation 1:30 How text knowledge graph representation works 2:08 💡 Force-atlas layout: why is it useful? 3:22 💡 Social network analogy 5:14 ❗ These are not vector word embeddings! 6:57 💡 Relevance ranking: betweenness centrality 8:54 How are topical clusters calculated? 10:07 💡 Extracting insights from the graph 11:11 ❗ Special feature: remove concepts to see the context around! 12:50 Interpreting AI Automation topics 13:36 💡 Generating a product idea from these insights 14:20 💡 Finding gaps between ideas to connect them in an interesting way 15:16 ❗Free business idea for you :) 16:27 The "blind spots" feature of InfraNodus 17:19 Use the built-in AI to generate product ideas 17:55 ❗Free app idea for you :) 18:46 BONUS CONTENT: Using the 3D graph (repetition is good for learning!) 19:11 ❗ Using InfraNodus with Obsidian folders! 20:25 Fast recap of the approach above using the Obsidian vault data To try: infranodus.com
Two questions: 1. Does it only use the search results as shown on the search results page or does it actually retrieve the content of each page and build the network from that? 2. Great point about the difference between embeddings and co-occurence, but what does co-occurrence mean in this case? Co-occurrence within a document, a paragraph, sentence, some kind of sliding window?
1. You can have both: use the titles and snippets selected by Google (and you can trust them as it's their business to retrieve the most relevant results) OR you can choose to import the pages' content 2. Co-occurrence within this particular text. If you'd like to see how it works, I wrote a peer-reviewed paper on the algorithm. You can find it here: dl.acm.org/doi/10.1145/3308558.3314123
Yes, this is because of the automatic lemmatizer, which is a bit aggressive. It doesn't change the meaning of your graph, but can make it look a bit funny. So you can set the language of your global user settings (or the graph) to English, and it won't happen. Hope this helps!
It's not on our own site, but through the payment portal that is integrated with Stripe. Also, these days everything that's https is safe. Finally, you can also pay via PayPal.
Timecodes:
0:00 Introduction
0:48 Content import: Google search results for AI automation
1:30 How text knowledge graph representation works
2:08 💡 Force-atlas layout: why is it useful?
3:22 💡 Social network analogy
5:14 ❗ These are not vector word embeddings!
6:57 💡 Relevance ranking: betweenness centrality
8:54 How are topical clusters calculated?
10:07 💡 Extracting insights from the graph
11:11 ❗ Special feature: remove concepts to see the context around!
12:50 Interpreting AI Automation topics
13:36 💡 Generating a product idea from these insights
14:20 💡 Finding gaps between ideas to connect them in an interesting way
15:16 ❗Free business idea for you :)
16:27 The "blind spots" feature of InfraNodus
17:19 Use the built-in AI to generate product ideas
17:55 ❗Free app idea for you :)
18:46 BONUS CONTENT: Using the 3D graph (repetition is good for learning!)
19:11 ❗ Using InfraNodus with Obsidian folders!
20:25 Fast recap of the approach above using the Obsidian vault data
To try: infranodus.com
Hi Dmitry, I love this demo, mixing scientific approach (graphs theory), illustrated with metahors, and marketing oriented examples. Great job !!
Thank you! Great to hear!
Thank you for sharing this video
Two questions: 1. Does it only use the search results as shown on the search results page or does it actually retrieve the content of each page and build the network from that? 2. Great point about the difference between embeddings and co-occurence, but what does co-occurrence mean in this case? Co-occurrence within a document, a paragraph, sentence, some kind of sliding window?
1. You can have both: use the titles and snippets selected by Google (and you can trust them as it's their business to retrieve the most relevant results) OR you can choose to import the pages' content
2. Co-occurrence within this particular text. If you'd like to see how it works, I wrote a peer-reviewed paper on the algorithm. You can find it here: dl.acm.org/doi/10.1145/3308558.3314123
I’ve noticed that some of the words are cut off in my maps. For example, “computer” might show up as “compu”. Can you help me understand why this is?
Yes, this is because of the automatic lemmatizer, which is a bit aggressive. It doesn't change the meaning of your graph, but can make it look a bit funny. So you can set the language of your global user settings (or the graph) to English, and it won't happen. Hope this helps!
I want to signup, but you are asking credit card on your own site, which is not safe. Can you please integrate stripe or something?
It's not on our own site, but through the payment portal that is integrated with Stripe. Also, these days everything that's https is safe. Finally, you can also pay via PayPal.