Getting the document-per-topic loading using TextmineR package by passing term co-occurrence matrix - r

Question

I am using TextmineR package to find the most similar documents to given document list. I used the following code to generate the tcm not dtm
tcm <- CreateTcm(doc_vec = text_df$Description,
skipgram_window = 20,
verbose = FALSE,
cpus = 2)
Which is used to fit a lda model:
# note the number of topics is arbitrary here
# see extensions for more info
model <- FitLdaModel(dtm = tcm,
k = 25,
iterations = 200, # I usually recommend at least 500 iterations or more
burnin = 180,
alpha = 0.1,
beta = 0.05,
optimize_alpha = TRUE,
calc_likelihood = TRUE,
calc_coherence = TRUE,
calc_r2 = TRUE,
cpus = 2)
Now the model parameter theta here generates word-per-topic loading rather than document-per-topic loading. I want to retrieve the document number from the document-per-topic loading. Please help in suggesting the method to obtain the document-per-topic distribution from this model while passing term co-occurrence matrix.
I have tried to back connect to get document number from document-per-topic loading, but not successful as per the guidelines given at https://cran.r-project.org/web/packages/textmineR/vignettes/d_text_embeddings.html

Answer 1

11-month old question. But giving it a shot anyway.
Technically, theta with LDA embeddings gives you P(topic|word) and phi still gives you P(word|topic). If I understand you correctly, you want to embed whole documents under this model? If so, here's how you'd do it.
library(textmineR)
# create a tcm
tcm <- CreateTcm(nih_sample$ABSTRACT_TEXT, skipgram_window = 10)
# fit an LDA model
m <- FitLdaModel(dtm = tcm, k = 100, iterations = 100, burnin = 75)
# pull your documents into a dtm
d <- nih_sample_dtm
# get them predicted under the model
# I recommend using the "dot" method for prediction with embeddings as sparsity may
# result in underflow and throw an error using the default "gibbs" method
p <- predict(object = m, newdata = d, method = "dot")