绘制LDA主题随时间的演变

时间:2015-07-28 15:32:21

标签: r ggplot2 tm lda topic-modeling

我想绘制一个特定主题的比例如何随时间变化,但我在隔离单个主题和绘制时间方面遇到了一些麻烦,尤其是分别绘制多组文档时(让我们创建两组进行比较 - 期刊A和B)。我在一个名为dateConverter的函数中保存了与这些日记相关联的日期。

这是我到目前为止所做的事情(非常感谢@scoa):

library(tm); library(topicmodels);


txtfolder <- "~/path/to/documents/"
source <- DirSource(txtfolder)

myCorpus <- Corpus(source, readerControl=list(reader=readPlain))


for (i in 1:10){
  meta(myCorpus[[i]], tag = "origin") <- "A"
}
for (i in 11:length(myCorpus)){
  meta(myCorpus[[i]], tag = "origin") <- "B"
}
dates <- do.call("c", dateConverter)
for (i in 1:length(myCorpus)){
  meta(myCorpus[[i]], tag = "datetimestamp") <- dates[i]
}

dtm <- DocumentTermMatrix(myCorpus, control = list(minWordLength=3))


n.topics <- 10
lda.model <- LDA(dtm, n.topics)
terms(lda.model,10)
df <- data.frame(id=names(topics(lda.model)),
                 topic=posterior(lda.model),
                 date=as.POSIXct(unlist(lapply(meta(myCorpus,type="local",tag="datetimestamp"),as.character))),
                 origin=unlist(meta(myCorpus,type="local",tag="origin"))    )

我如何绘制这些?

1 个答案:

答案 0 :(得分:5)

这只是我之前回答的改编:

## Load the data
library(tm)

## Use built-in data set
data(acq)
myCorpus <- acq

## prepare the data
for (i in 1:25){
  meta(myCorpus[[i]], tag = "origin") <- "A"
}

for (i in 26:length(myCorpus)){
  meta(myCorpus[[i]], tag = "origin") <- "B"
}

dates <- sample(seq.Date(as.Date("2013-01-01"),as.Date("2014-01-01"),length.out=8),50, replace=TRUE)

for (i in 1:length(myCorpus)){
  meta(myCorpus[[i]], tag = "datetimestamp") <- dates[i]
}

dtm <- DocumentTermMatrix(myCorpus, control = list(minWordLength=3))


library(topicmodels)    
n.topics <- 5
lda.model <- LDA(dtm, n.topics)
terms(lda.model,10)

重塑数据以进行绘图。我对每组主题,日期和起源采取平均后验。

df <- data.frame(id=names(topics(lda.model)),                 
                 date=as.POSIXct(unlist(lapply(meta(myCorpus,type="local",tag="datetimestamp"),as.character))),
                 origin=unlist(meta(myCorpus,type="local",tag="origin"))    )

dft <- cbind(df,posterior(lda.model)$topics)

library(dplyr)
library(tidyr)
M <- gather(dft,topic,value,-id,-date,-origin) %>%
  group_by(topic,date,origin) %>%
  summarize(value=mean(value))

剧情

library(ggplot2)
ggplot(M,aes(x=date,color=origin,y=value)) + 
  geom_point() +
  geom_line() +
  facet_grid(topic~origin)

enter image description here