back to list

Project: Dynamic Knowledge Graph Embeddings

Description

Knowledge graph embeddings are an important area of research inside machine learning and has become a necessity due to the importance of reasoning about objects, their attributes and relations in large graphs. There have been several approaches that have been explored and can be categorized based on the underlying approaches: matrix factorization, deep learning, edge reconstruction and graph kernels. Most of the research in this field has limited itself to the study of static graphs thereby ignoring the fact that majority of the real world data sets or knowledge graphs are dynamic in nature i.e. they change/evolve over the course of time. Even when such dynamic graphs are considered, they are treated as a set of static graphs so that existing techniques can be applied. In this project we propose to work on dynamic graphs that can capture such evolving graphs naturally inside a graph neural network framework. This can help us harbor the expressive power of graph neural networks for several real world problems while also taking advantage of the structure of these evolving graphs. To this effect, we propose to come up with a dynamic graph neural network model and apply it to various real-world knowledge graphs.

References:
[1] Wu et al., Efficiently embedding dynamic knowledge graphs, Knowledge-based Systems, 2022
[2] https://github.com/woojeongjin/dynamic-KG

Details

Supervisor

Devendra Dhami

Interested?

Get in contact