TWed Talk: Sola Shirai on "An Introduction to Rule-Learning Models for Link Prediction" 20 Oct 2022

Link prediction methods can be a valuable tool to identify new or missing facts in a knowledge graph, and in research involving the use of KGs it also can serve as a baseline method to perform tasks like item recommendation or disease diagnosis. Among the many models that can perform this task, rule-based models can be particularly useful, as they provide an inherent level of explainability and can more easily be applied to previously-unseen entities. In this TWed, we will first discuss the task of link prediction in general -- what is the task, how it might be useful to you, and what are its limitations -- followed by an in-depth look into how some rule-based models actually learn and apply rules in order to make predictions.