Knowledge Graphs & Deep Learning at YouTube

Learn how can you combine Knowledge Graphs and Deep Learning to dramatically improve Search & Discovery systems, just like YouTube does. By using a combination of signals (audiovisual content, title & description and context), it is possible to find the main topics of a video. These topics can then be used to improve collaborative filtering, search, structured browsing (by exploiting the structure of the knowledge graph), ads, and much more.

Slides: https://www.slideshare.net/aurelienge...

Filmed at https://2018.dotai.io on May 31st in Paris. Big thanks to the Dot AI team! More talks on https://www.dotconferences.com/talks

Google is very open about its research. Everything I talked about in this talk (and much more!) is available in many papers, blogs and talks. Here are a few pointers if you want to learn more:
"Deep Neural Networks for YouTube Recommendations" paper by Paul Covington, Jay Adams and Emre Sargin (2016): https://research.google.com/pubs/arch...
"Large-scale Video Classification with Convolutional Neural Networks", Karpathy et al.(2014): https://cs.stanford.edu/people/karpat...
Google I/O 2013 on YouTube video annotations:    • Google I/O 2013 - Semantic Video Annotatio...  
"Classifying YouTube Channels: a Practical System", Vincent Simonet (2013): https://research.google.com/pubs/arch...
Wikidata: https://wikidata.org/
Java framework for building Semantic Web and Linked Data applications: https://jena.apache.org/
SPARQL tutorial: https://jena.apache.org/tutorials/spa...
List of papers on Entity Recognition using Deep Learning: https://memkite.com/deeplearningkit/2...
Extracting information from text using the NLTK library (python): https://www.nltk.org/book/ch07.html
Project that lists Entity Linking tools and evaluates them on various datasets: http://aksw.org/Projects/GERBIL.html

Kudos to the YouTube Legos team in Paris, the VCA team in Mountainview, the YouTube Recommendation team in San Bruno, and to all the other teams I had such pleasure working with at YouTube and Google! :)