Collaborative Topic Models for Users and Texts

Chong Wang, senior research scientist at Baidu, presents his research on probabilistic topic models.

Probabilistic topic models provide useful tools to analyze large-scale text corpora by discovering underlying hidden themes. We can use these models to visualize, summarize and navigate document collections. Nowadays, many document collections are associated with reader behavior data and we are interested in using these data to understand patterns in how they read and make recommendations for the readers.

In this talk, Chong reviews the basics of topic modeling and describes collaborative topic models---models that simultaneously model documents and users'  reading behaviors. These models can be used to discover patterns in how people read / how documents are read; and those patterns, in turn, can help readers to find relevant new documents. He also present a simple demo that incorporates some of these ideas into a recommender system for scientific articles.

This video was recorded at the SF Bayarea Machine Learning meetup at Galvanize.