Yi Zhang 

 

Date: Wednesday May 1, 2019

Time: 1:00pm – 2:00pm 

Location: Essex Hall Room 109

 

 

Representing networks is crucial for many downstream tasks, such as classification, clustering, and link prediction. The success of Skip-gram model (hereafter SGNS) has inspired many network embedding algorithms, such as LINE, DeepWalk, and node2vec. In this seminar, we first show that these algorithms suffer from the norm convergence problem, then propose to use L2 regularization to rectify the problem. The proposed method improves the embeddings consistently on different datasets with various sizes and structures. Another problem is that existing methods cannot scale to large networks. In our work, we propose N2V, a random walk based online sampling method to learn the network embeddings. Experiments show that N2V can learn embeddings from large network effectively. It also has good performance on node classification, link prediction and author clustering tasks.

 

Thesis Committee: 

Seminar Announcement

5113 Lambton Tower, 401 Sunset Ave., Windsor, ON, N9B 3P4, (519) 253-3000 ext. 3716, csgradinfo@uwindsor.ca