Hyperbolic Metric Learning Embedding for User Behavior Sequence
Tse-Yu Lin · 2024
The sequential behavior history is an important data in the field of consumer electronics, A common way is to model all browser history as a graph, thus, each sequence of user behavior can be view as a walk on the certain graph. Further, we could model individual user’s historical behavior as a tree sub-graph. Recently, tree-structure data is widely studied, and various hyperbolic metric learning methods are proposed to find a nice representation into a hyperbolic space for the usage of downstream tasks. In this work, we propose a framework to deal with sequential behavior history defined by tags of website pages, then apply hyperbolic metric learning models to obtain an appropriate representation each user.