Typology of mixed-membership models: towards a design method

Authors:
Gregor Heinrich
Affiliations:
University of Leipzig + vsonix GmbH, Darmstadt, Germany
Venue:
ECML PKDD'11 Proceedings of the 2011 European conference on Machine learning and knowledge discovery in databases - Volume Part II
Year:
2011

Citing 12
Cited 1

Latent dirichlet allocation

The Journal of Machine Learning Research
Matching words and pictures

The Journal of Machine Learning Research
The author-topic model for authors and documents

UAI '04 Proceedings of the 20th conference on Uncertainty in artificial intelligence
Pachinko allocation: DAG-structured mixture models of topic correlations

ICML '06 Proceedings of the 23rd international conference on Machine learning
Statistical entity-topic models

Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining
Latent Dirichlet Co-Clustering

ICDM '06 Proceedings of the Sixth International Conference on Data Mining
Mixtures of hierarchical topics with Pachinko allocation

Proceedings of the 24th international conference on Machine learning
Modeling online reviews with multi-grain topic models

Proceedings of the 17th international conference on World Wide Web
Clustering the tagged web

Proceedings of the Second ACM International Conference on Web Search and Data Mining
A Generic Approach to Topic Models

ECML PKDD '09 Proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases: Part I
Multi-HDP: a non parametric Bayesian model for tensor factorization

AAAI'08 Proceedings of the 23rd national conference on Artificial intelligence - Volume 3
Variational Bayes for generic topic models

KI'09 Proceedings of the 32nd annual German conference on Advances in artificial intelligence

Sweeping through the topic space: bad luck? Roll again!

ROBUS-UNSUP '12 Proceedings of the Joint Workshop on Unsupervised and Semi-Supervised Learning in NLP

Quantified Score

Hi-index	0.00

Visualization

Abstract

Presents an analysis of the structure of mixed-membership models into elementary blocks and their numerical properties. By associating such model structures with structures known or assumed in the data, we propose how models can be constructed in a controlled way, using the numerical properties of data likelihood and Gibbs full conditionals as predictors of model behavior. To illustrate this "bottom-up" design method, example models are constructed that may be used for expertise finding from labeled data.