Stanford CS224W: Machine Learning with Graphs | 2021 | Lecture 11.3 – Query2box: Reasoning over KGs

For more information about Stanford’s Artificial Intelligence professional and graduate programs, visit: https://stanford.io/3bngZHH

Lecture 11.3 – Query2box Reasoning over KGs Using Box Embeddings
Jure Leskovec
Computer Science, PhD

In this video, we show how to answer more complex queries with complex conjunction operations over KGs. We show how box embeddings can be used to embed queries and how to use the intersection of these boxes to answer conjugation of queries. We then talk about how to learn entity and relationship embeddings and define an intersection operator using a geometric interpretation. We also define the notion of distance between boxes and entities. Finally, we show how to also answer more general conjunctive-disjunctive queries which take the form of AND-OR and show how we can learn all our KG model parameters by optimizing a loss function.

To follow along with the course schedule and syllabus, visit:
http://web.stanford.edu/class/cs224w/