align_topics: Align topics from distinct LDA models

In lasy/alto: Aligns Topics across LDA Models

Align topics from distinct LDA models

Description

This function takes a list of LDA models and returns an object of class alignment. Each element in the models list must be itself a named list, corresponding to the mixed memberships ($gamma) and topics ($beta). The resulting alignment object can be plotted using 'plot' and its weights can be extracted using the 'weights' accessor function. See the documentation for class alignment for further details.

Usage

align_topics(models, method = "product", ...)

Arguments

models	(required) A list of LDA models object. Each list component must be a list with two named entries, $gamma (containing mixed memberships) and $beta (containing topic parameters in log sapce). See run_lda_models for details.
method	(required) Either product or transport, giving two types of alignment strategies, using inner products between gamma vectors or optimal transport between gamma-beta pairs, respectively. Defaults to product.
...	(optional) Further keyword arguments passed to the weight function. For example, passing reg = 10 when using the transport method will use a regularization level to 10 in the Sinkhorn optimal transport algorithm.

Details

After topics are aligned, they are re-ordered such that topics connected by high weights are ranked similarly within their respective models.

Topic paths (sets of topics connected by high weights across models) are then identified and alignment diagnostics (topic refinement and coherence scores) are computed. These variables are included to the topics container of the returned alignment.

Value

An object of class alignment providing the weights between every pair of topics of each model pairs in the input edgelist (comparisons).

See Also

alignment

Examples

library(purrr)
data <- rmultinom(10, 20, rep(0.1, 20))
lda_params <- setNames(map(1:5, ~ list(k = .)), 1:5)
lda_models <- run_lda_models(data, lda_params)

alignment <- align_topics(lda_models)
alignment
plot(alignment)

plot(alignment, color_by = "refinement")
alignment <- align_topics(lda_models, method = "transport")
plot(alignment)
plot_beta(alignment)

topics(alignment)
weights(alignment)
models(alignment)

lasy/alto documentation built on June 23, 2024, 6:45 a.m.