How to perform Structural Topic Modeling (stm) in R? This code works on Twitter Data but can be used by any corpus if it has a unique id field

Data Pipleline for Structural Topic Modeling in R for Twitter Data

library(tidyverse)
loc <- "FILEPLATH /data.csv"
tweets <- read_csv(loc) # fetching data

# Data is already processed in Python but it can also be done in R
# Data Processing steps:
# All lower cased, No URLS, No stop words, No punctuations but # and @, 

library(quanteda) 

tweetCorpus <- corpus(tweets$body) # creating corpus
docvars(tweetCorpus, field = "cov1") <- tweets$local # covariance field local
docvars(tweetCorpus, field = "cov2") <- tweets$HotIssue # covariance field HotIssue
docvars(tweetCorpus, "id") <- tweets$tweetId # adding this field to keep track of tweet ids
# each date is converted into a number
docvars(tweetCorpus, "cov3Time") <- as.integer(as.factor(substr(tweets$postedTime,1,10))) # covariance field postedtime

# add words that you think are stopwords based on your corpus as well
# Use term document frequency results to identify those
stopWords <- c("t.co","http","https","amp","t","t.c","c","rt") # replacing stopwords if any

# Converting corpus to document frequency matrix
tweetDfm <- dfm(tweetCorpus,
           remove = c(stopwords("english"), stopWords), # replacing all those stop words
           ngrams= 1L,
           stem = F, # No stemming
           remove_numbers = TRUE,  
           remove_punct = TRUE, 
           remove_symbols = TRUE) %>%
  dfm_trim(min_count = 10, min_docfreq = 5) 
# min_count = remove words used less than x
# min_docfreq = remove words used in less than x docs

# showing top 50 words with highest frequency
topfeatures(tweetDfm, n = 50)

library(stm)

# converting dfm to stmdfm
stmtweetDfm <- convert(tweetDfm, to = "stm")

# preparing out variable for ctm or stm model
out <- prepDocuments(stmtweetDfm$documents, 
                               stmtweetDfm$vocab, 
                               stmtweetDfm$meta, 
                     lower.thresh = 10) # 


tweetsRemoved <- stmtweetDfm$meta$id[-out$docs.removed] # fetching tweet ids of all documents not removed
tweetsCleaned <- subset(tweets,tweets$tweetId %in% tweetsRemoved) # fetching tweets that were not removed
save(tweetsCleaned, file = "FILEPATH/tweetsCleaned.RData")


k <- 40 # number of topics
stmResult <- stm(out$documents, out$vocab, K = k,prevalence = ~cov1 + cov2 + s(cov3Time),
                          max.em.its = 180, data = out$meta, init.type = "Spectral", seed = 300)

save(stmResult, file = "FILEPATH/stmResult.Rdata")