Module 26: Embedding a Network into Latent Space

Faculty Contact: Ambuj Singh

Research Area(s): 

Abstract: This module will focus on understanding how to embed a network of Twitter users into latent space, and explore network dynamics on the embedded network. In the original network, each user will correspond to a node, with directed edges ( u, v) between users if user u follows user v. In order to find a latent space representation for each user, this module will experiment with different embeddings based on topic modeling and other dimensionality reduction techniques.

This module will then measure the homophily of each embedded graph in order to determine which embeddings best represent the original network. Ideally, users who are linked in the original network will be “close” together in latent space.

Finally, users will be clustered according to their position in latent space, in order to capture groups of users with similar attributes and behaviors. These clusters may then be used to model cascading behavior in the network - for instance, when a user’s behavior such as retweeting may cause another user to retweet or like a post.

Active Quarters:

  • Winter 2017: Rachel Redberg