Mike Tung on lessons learned from building, maintaining, and using a knowledge graph with over a trillion connected facts.
Mike Tung is founder and CEO of Diffbot, a startup that crawls the web and provides one of the most comprehensive knowledge graphs, accessible through a variety of simple interfaces. Although knowledge graphs can be described in a variety of ways, an article from 2016 suggested the following succinct framing:
- A knowledge graph acquires and integrates information into an ontology and applies a reasoner to derive new knowledge.
We discussed applications of knowledge graphs in search, recommendations, as features in AI and machine learning applications, and as core components in AI systems capable of blending ML models with knowledge.
Google really first popularized that word with the Google Knowledge Graph. This was basically a startup they acquired called Metaweb which had this knowledge graph called Freebase. When you search for something on Google, on the right hand side, you see these knowledge panels. So that’s what people refer to knowledge graphs in search – they’re referring to this information in a knowledge panel.
Many major companies now have knowledge graphs, Spotify, Netflix, Uber all have Knowledge Graph teams. When they are doing search and recommendations within their apps, like when you open up the Uber Eats app, for example, it’s recommending dishes and restaurants for you. In this example Uber is pulling that from a backend knowledge graph where they have modeled out all the dishes and all the restaurants and in specific locations. Using that information when doing ranking is very natural and produces better and also more explainable recommendations.
Highlights in the video version:
- Introduction to Mike Tung, Founder and CEO of Diffbot
Knowledge Graph and Diffbot Knowledge Graph
Knowledge Graph and storing information
Use cases where Knowledge Graph shines
Advantage of using Diffbot and Knowledge Graph
Augment models using Diffbot
Key facets when evaluating Knowledge Graph
Top case studies of how Diffbot is being used
Who are Diffbot users?
What is the profile of the application?
What are common ways Diffbot data is used for ML?
Impact of NLP, transformers, and language models
Infusing knowledge to build ML Systems has exploded
Explainability in AI, neural networks, and deep neural networks
Academics are more data centric AI
Blending domain and human knowledge
What is the Knowledge Graph community?
Non English languages represented in Diffbot
Industry trends what will impact Diffbot in a year or two
Breakthrough models that are cited all over the place
When will usable, accessible, and multimodel models be available?
What’s next for Diffbot? What are you hiring for?
Knowledge Graphs and NLP
Images, videos, audio, podcasts, texts, and embeddings
An mp3 file should be transcribed and analyzed in all modes
- A video version of this conversation is available on our YouTube channel.
- “What is Graph Intelligence?”
- Mayank Kejriwal: “Building and deploying knowledge graphs”
- Denise Gosnell: “How graph technologies are being used to solve complex business problems”
- “Top Places to Work for Data Engineers”
- Azeem Ahmed: “Data and Machine Learning Platforms at Shopify”
- Che Sharma: “Modern Experimentation Platforms”
Download the FREE report below:
[Image: Knowledge Graph of the public web from Diffbot]