Topic Models in Machine Learning and Cognitive Science
Dr. Mark Steyvers
Tuesday, December 02, 2014, 01:00pm - 02:00pm
University of California, Irvine, Department of Cognitive Sciences
Probabilistic topic models are widely used in machine learning to analyze the content of unstructured text from newspaper articles, blogs, email and dialogue. They are also used in cognitive science as a computational analog of how human learners might discover semantic knowledge through their linguistic experience. Part I of the talk will focus on using topic models for the purpose of exploratory data analysis. I will present the basic topic model, Latent Dirichlet Allocation (LDA) as well as a number of variants that have been introduced such Labeled LDA, a machine learning approach for multiple-label document classification. This model allows us to analyze the connections between individual words in documents and sets of content labels associated with documents. I demonstrate how to model can be used to automatically label psychotherapy transcripts according to the subject of conversation as well as the symptoms displayed by the patient. Part II of the talk will focus on the topic model as a model for human memory. I will present a version of a topic model that combines gist-based representations and context-specific episodic memories. I show how this probabilistic memory model can account for the influence of prior knowledge in episodic memory tasks.
Note: If you would like to receive email announcements about the colloquium series, please contact the Business Office to have your name added to our announce lists at firstname.lastname@example.org.