Difference between revisions of "Topic model"
From Cohen Courses
Jump to navigationJump to searchm |
(No difference)
|
Revision as of 16:43, 15 March 2010
This is a technical method discussed in Social Media Analysis 10-802 in Spring 2010.
Topic models are probabilistic, generative models that use Hierarchical Bayesian Analysis of a document collection (data) to uncover the underlying semantic structure. Different types of topic models have been proposed to capture different notions of the underlying semantic structure, including :
- A document may have words from multiple topics (LDA)
- Links between documents imply homophily in topic-space (link-LDA, Relational topic model)
- Utilizing known labels on documents while learning parameters (supervised-LDA)
- Authors write similar documents (Author-Topic model)
- Authors write similar documents, conditioned on who the recipient is. (Author-Recipient-Topic model)