Date:  April 15, 2020

Time: 11:00 am-12:30am

Zoom Meeting URL: https://zoom.us/j/429428145?pwd=VEF4L0tqSklkVlZ4UE5WWlkrYWtwQT09

Meeting ID: 429 428 145

Password: Request from csgradinfo@uwindsor.ca

 

In scholarly data, authors are connected via collaboration and further linked by citation links between papers, forming a heterogeneous network that contains richer information than homogeneous networks. Learning the embedding for authors is a crucial task for analyzing authors. 

 

In most network embedding algorithms, long random walks are often used to convert the graph into `text' and node embeddings can be learned by Skip-gram with Negative Sampling (SGNS) model. The state-of-the-art algorithm is DeepWalk. However, academic networks are usually directed graphs, where long random walks can be trapped or interrupted, leading to low-quality embeddings. In our work, we use a directed network embedding method, called ShortWalk, to learn author embeddings on directed graphs. ShortWalk generates short random traces and gives nodes equal weight by using the pair-wise combination to generate the training pairs. We apply ShortWalk on the heterogeneous author-paper network to learn author embeddings, and experiments show that ShortWalk performs better than DeepWalk in the author classification task.

 

Thesis Committee: 

 

PhD Seminar Announcement