My Thoughts on Google’s Knowledge Graph
Talking about this announcement, there is nothing really new, but it is an interesting way of telling the story about Google’s competition with network of data stored in Facebook.
The biggest problem is that graphs Google & Facebook have are different, lets compare them:
- Google has a network of connections between pages, but not between real things (PageRank grew out of links between scientific articles)
- Facebook, in contrast, has network of connections between people, and also network of connections between people and web pages they reference to
Now, Facebook has places and other elements like cities, products, companies, etc., and also has connections between them and people.
- Also, as places have location stored (in form of physical address of latlong), Facebook also has a geographic data useful to show related places based on location (show me restaurants near Nikolina Gora), which adds a geo dimension to data.
- Then, Facebook also has a database of check-ins (ok, granted, 4sq might have a larger one, still, Facebook has larger userbase), so it has a connection between people, places and time they visited these places.
- So, Facebook has a lot of knowledge captured in form of entities manually added and checked for content quality (and owners in several cases – like for products, companies, etc.), while Google has sophisticated entity extraction technologies to do the same job that Facebook users already made.
- At the end of the day, all connections are created by people. This is important.
- Facebook beats Google with its ability to have entities created using a combination of human and simple entity extraction while Google largely uses sophisticated entity extraction approach.
- As it was commented in the article, computers will bring only their own understanding and point of view.
- I agree with importance of automatic data extraction, believe into Open Graph initiative by Facebook, but I also believe that combination of human and AI will better serve the Knowledge graph that any AI-only initiative brought to life by Google.
- Personally, I believe that human role in data organization will be still the leading power of information organization in the XXI century, and computers will be in best cases advising us.
Zet Universe and Interest Graph
Coming back to Zet Universe, we are using an approach more similar to the one Facebook has. We provide users with tools to organize their knowledge around their activities. We look into data and provide search on it. We use semantic indexing algorithms to find similar documents by the content, but we record user interaction with content, and we record positions of elements stored on the infinite zoomable space. We provide lightweight semantic markup tools to users to help them to mark/tag information as they think about it.
This helps us to understand similarity & identify connections between entities by following our users thoughts. And this is the way we build our Interest Graph (as we call it).
We believe that computers shouldn’t make visualization of our own knowledge automatically but rather give us a chance to do it in the way we think about it.
To sum our approach, here is a short statement:
It is good idea to automate visualization for data I have no sense for, but for my data I want to see it organized in the way I think about it.