On smoothing and inference for topic models
Latent Dirichlet analysis, or topic modeling, is a flexible latent variable framework for modeling high-dimensional sparse count data. Various learning algorithms have been developed in recent years, including collapsed Gibbs sampling, variational inference, and maximum a posteriori estimation, and...
| Main Authors: | , , , |
|---|---|
| Format: | Journal article |
| Sprog: | English |
| Udgivet: |
2009
|
| _version_ | 1826295130728955904 |
|---|---|
| author | Asuncion, A Welling, M Smyth, P Teh, Y |
| author_facet | Asuncion, A Welling, M Smyth, P Teh, Y |
| author_sort | Asuncion, A |
| collection | OXFORD |
| description | Latent Dirichlet analysis, or topic modeling, is a flexible latent variable framework for modeling high-dimensional sparse count data. Various learning algorithms have been developed in recent years, including collapsed Gibbs sampling, variational inference, and maximum a posteriori estimation, and this variety motivates the need for careful empirical comparisons. In this paper, we highlight the close connections between these approaches. We find that the main differences are attributable to the amount of smoothing applied to the counts. When the hyperparameters are optimized, the differences in performance among the algorithms diminish significantly. The ability of these algorithms to achieve solutions of comparable accuracy gives us the freedom to select computationally efficient approaches. Using the insights gained from this comparative study, we show how accurate topic models can be learned in several seconds on text corpora with thousands of documents. |
| first_indexed | 2024-03-07T03:56:20Z |
| format | Journal article |
| id | oxford-uuid:c2fd0c46-213d-4dce-a6ae-4b0693ee761e |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T03:56:20Z |
| publishDate | 2009 |
| record_format | dspace |
| spelling | oxford-uuid:c2fd0c46-213d-4dce-a6ae-4b0693ee761e2022-03-27T06:13:09ZOn smoothing and inference for topic modelsJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:c2fd0c46-213d-4dce-a6ae-4b0693ee761eEnglishSymplectic Elements at Oxford2009Asuncion, AWelling, MSmyth, PTeh, YLatent Dirichlet analysis, or topic modeling, is a flexible latent variable framework for modeling high-dimensional sparse count data. Various learning algorithms have been developed in recent years, including collapsed Gibbs sampling, variational inference, and maximum a posteriori estimation, and this variety motivates the need for careful empirical comparisons. In this paper, we highlight the close connections between these approaches. We find that the main differences are attributable to the amount of smoothing applied to the counts. When the hyperparameters are optimized, the differences in performance among the algorithms diminish significantly. The ability of these algorithms to achieve solutions of comparable accuracy gives us the freedom to select computationally efficient approaches. Using the insights gained from this comparative study, we show how accurate topic models can be learned in several seconds on text corpora with thousands of documents. |
| spellingShingle | Asuncion, A Welling, M Smyth, P Teh, Y On smoothing and inference for topic models |
| title | On smoothing and inference for topic models |
| title_full | On smoothing and inference for topic models |
| title_fullStr | On smoothing and inference for topic models |
| title_full_unstemmed | On smoothing and inference for topic models |
| title_short | On smoothing and inference for topic models |
| title_sort | on smoothing and inference for topic models |
| work_keys_str_mv | AT asunciona onsmoothingandinferencefortopicmodels AT wellingm onsmoothingandinferencefortopicmodels AT smythp onsmoothingandinferencefortopicmodels AT tehy onsmoothingandinferencefortopicmodels |