Install package from github (only once for the first time). Remember that this is first public version of the package and it can contain errors.

devtools::install_github("nikita-moor/ldatuning")

Package ldatuning realize 4 metrics to select perfect number of topics for LDA model.

library("ldatuning")

Load “AssociatedPress” dataset from the topicmodels package.

library("topicmodels")
data("AssociatedPress", package="topicmodels")
dtm <- AssociatedPress[1:10, ]

The most easy way is to calculate all metrics at once. All existing methods require to train multiple LDA models to select one with the best performance. It is computation intensive procedure and ldatuning use parallelism, so do not forget to point correct number of CPU cores in mc.core parameter to archive the best performance. This version of package uses mclapply which doesn’t actually run in parallel on Windows (reverts to serial evaluation), we are going to use doParallel or some other cross-platform method in future version.

All standard LDA methods and parameters from topimodels package can be set with method and control.

result <- FindTopicsNumber(
  dtm,
  topics = seq(from = 2, to = 15, by = 1),
  metrics = c("Griffiths2004", "CaoJuan2009", "Arun2010", "Deveaud2014"),
  method = "Gibbs",
  control = list(seed = 77),
  mc.cores = 2L,
  verbose = TRUE
)
## fit models... done.
## calculate metrics:
##   Griffiths2004... done.
##   CaoJuan2009... done.
##   Arun2010... done.
##   Deveaud2014... done.

Result is a number of topics and corresponding values of metrics

topics Griffiths2004 CaoJuan2009 Arun2010 Deveaud2014
2 -15349.79 0.1169522 9.888687 0.6989189
3 -15266.66 0.1600736 8.959751 0.5318997
4 -15242.86 0.1779016 8.535274 0.4323482
5 -15226.91 0.1875260 7.942649 0.3718687
6 -15251.04 0.2612029 7.904418 0.3101625
7 -15259.80 0.2746812 7.621045 0.2746203
8 -15256.30 0.3061726 7.591150 0.2435689
9 -15303.87 0.3379840 7.550364 0.2181424
10 -15291.00 0.3829542 7.689896 0.1969989
11 -15293.55 0.4347111 7.873546 0.1770861
12 -15326.94 0.4756351 8.237908 0.1594651
13 -15319.82 0.4944709 8.236753 0.1504368
14 -15338.24 0.4927860 7.965343 0.1406462
15 -15297.82 0.5047240 8.316559 0.1362596

Simple approach in analyze of metrics is to find extremum, more complete description is in corresponding papers:

For easy analyze of the results can be used support function FindTopicsNumber_plot

FindTopicsNumber_plot(result)

Results calculated on the whole dataset (about 10 hours on quad-core computer) looks like

From this plot can be made conclusion that optimal number of topics is in range 90-140. Metric Deveaud2014 is not informative in this situation.

References

1.Rajkumar Arun, V. Suresh, C. E. Veni Madhavan, and M. N. Narasimha Murthy. 2010. On Finding the Natural Number of Topics with Latent Dirichlet Allocation: Some Observations. In Advances in Knowledge Discovery and Data Mining, Mohammed J. Zaki, Jeffrey Xu Yu, Balaraman Ravindran and Vikram Pudi (eds.). Springer Berlin Heidelberg, 391–402. http://doi.org/10.1007/978-3-642-13657-3_43

2.Cao Juan, Xia Tian, Li Jintao, Zhang Yongdong, and Tang Sheng. 2009. A density-based method for adaptive LDA model selection. Neurocomputing — 16th European Symposium on Artificial Neural Networks 2008 72, 7–9: 1775–1781. http://doi.org/10.1016/j.neucom.2008.06.011

3.Romain Deveaud, Éric SanJuan, and Patrice Bellot. 2014. Accurate and effective latent concept modeling for ad hoc information retrieval. Document numérique 17, 1: 61–84. http://doi.org/10.3166/dn.17.1.61-84

4.Thomas L. Griffiths and Mark Steyvers. 2004. Finding scientific topics. Proceedings of the National Academy of Sciences 101, suppl 1: 5228–5235. http://doi.org/10.1073/pnas.0307752101

5.Martin Ponweiser. 2012. Latent Dirichlet Allocation in R. Retrieved from http://epub.wu.ac.at/id/eprint/3558