How to output the themes with time trends about rollinglda

[JonasRieger]

There is no built-in package function for things like this. But it is quite easy to extract such information from the object:

library(rollinglda)

# roll = rollinglda object
dates = getDates(roll)

# to extract monthly topic matrices:
months = lubridate::floor_date(dates, "month")
xmonth = sort(unique(months))

topicsm = lapply(xmonth, function(x){
  tmp = table(factor(unlist(getAssignments(getLDA(roll))[months == x])+1, levels = 1:getK(getLDA(roll))), 
              factor(unlist(lapply(getDocs(roll)[months == x], function(y) y[1,]))+1,
                     levels = seq_len(length(getVocab(roll)))))
  tmp = matrix(as.integer(tmp), nrow = getK(getLDA(roll)))
  colnames(tmp) = getVocab(roll)
  tmp
})

# then, you can, e.g., extract monthly topWords
topwordsm = lapply(topicsm, tosca::topWords, numWords = 50)

Maybe I'll include such a function for postprocessing in a future version of rollinglda. It feels quite convenient.

If you have any further questions, feel free to reach out.

[reve0913]

Thank you so much for your explanation.