en
0.25
0.5
0.75
1.25
1.5
1.75
2
Kernel Topic Models
Published on Jan 16, 20133580 Views
Latent Dirichlet Allocation models discrete data as a mixture of discrete distributions, using Dirichlet beliefs over the mixture weights. We study a variation of this concept, in which the documents'
Related categories
Chapter list
Kernel Topic Models00:00
Overview00:26
Introduction and Motivation01:15
Topic Models01:16
Topic Models and Metadata02:21
Documents in Context02:57
Related Work03:46
The Kernel Topic Model04:35
Topic Model as Matrix Factorization04:36
Conditional Topic Model05:22
Kernel Topic Model06:02
Latent Dirichlet Analysis06:24
Kernel Topic Model06:42
Kernels/Covariance Functions on Documents07:33
The Laplace Bridge: From Hilbert Space to Probabilities and back08:36
Kernel Topic Model: Laplace Bridge08:47
Inference: The Laplace Bridge09:28
Laplace approximation: softmax basis11:12
From Dirichlet to Gaussian (and back)13:00
Beta distributions approximated in Softmax basis14:30
Gaussian “Dirichlet” in Action15:32
Inference: Laplace Bridge vs MCMC16:33
Experimental Results17:32
State of the Union17:43
State of the Union: Kernel Topic Model21:20
State of the Union: Linear Model23:14
State of the Union: Perplexity23:59
More Perplexity24:56
Conclusions26:31