R: wordcloud包从语料中省略三个字符以下的单词

Question

When creating a wordcloud using the wordcloud package it seems like the package defaults to omitting words below three characters (such as "tv"). I think this is a feature rather than a bug, but still I could not find an argument that adjusts the minimum character count.

The wordcloud is run against a corpus of words created and preprocessed with the Corpus() and tm_map() functions from the tm package. I have confirmed that the words in question have not gotten lost when eg removing stopwords - they are still in the final corpus on which the wordcloud() function is run.

Reproducible example [edit]

Real data obviously looks different. However, the lines below replicates the error.

customPalette <- brewer.pal(4, "Dark2")

wordVector <- c(rep("tv", 15), rep("computer", 4), rep("phone", 16), rep("tablet",10))
newCorpus <- Corpus(VectorSource(wordVector))

wordcloud(newCorpus, max.words = 100, scale=c(8,1), random.order = FALSE, random.color = TRUE, colors = customPalette)

This creates output:

Session info:

R version 3.3.2 (2016-10-31)
Platform: x86_64-redhat-linux-gnu (64-bit)
Running under: Amazon Linux AMI 2016.09

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C               LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8     LC_MONETARY=en_US.UTF-8   
 [6] LC_MESSAGES=en_US.UTF-8    LC_PAPER=en_US.UTF-8       LC_NAME=C                  LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
 [1] bindrcpp_0.2             zoo_1.8-0                wordcloud_2.5            RColorBrewer_1.1-2       SnowballC_0.5.1          tm_0.7-1                
 [7] NLP_0.1-10               reshape2_1.4.2           lubridate_1.6.0          scales_0.4.1             ggplot2_2.2.1            aws.s3_0.3.3            
[13] githubinstall_0.2.1.9001 aws.signature_0.3.2      RJDBC_0.2-5              rJava_0.9-8              DBI_0.7                  RCurl_1.95-4.8          
[19] bitops_1.0-6             jsonlite_1.5             dplyr_0.7.0              sparklyr_0.5.6           drat_0.1.2               devtools_1.13.2         

loaded via a namespace (and not attached):
 [1] slam_0.1-40       lattice_0.20-34   colorspace_1.3-2  htmltools_0.3.6   yaml_2.1.14       base64enc_0.1-3   rlang_0.1.1       glue_1.1.1       
 [9] withr_1.0.2       dbplyr_1.0.0      bindr_0.1         plyr_1.8.4        stringr_1.2.0     munsell_0.4.3     gtable_0.2.0      memoise_1.1.0    
[17] labeling_0.3      httpuv_1.3.3      parallel_3.3.2    curl_2.6          Rcpp_0.12.11      xtable_1.8-2      backports_1.1.0   config_0.2       
[25] mime_0.5          digest_0.6.12     stringi_1.1.5     shiny_1.0.3       rprojroot_1.2     grid_3.3.2        tools_3.3.2       magrittr_1.5     
[33] lazyeval_0.2.0    tibble_1.3.3      pkgconfig_2.0.1   data.table_1.10.4 xml2_1.1.1        assertthat_0.2.0  httr_1.2.1        rstudioapi_0.6   
[41] R6_2.2.2          git2r_0.18.0

Answer 1

The problem appears whether using the vector wordVector or with the corpus version. This seems to be intended behaviour - see comment below from the package maintainer.

The following alternative approach works, using the ability of wordcloud to take a vector of words and their frequencies separately...

worddf <- as.data.frame(table(newCorpus$content))
wordcloud(words = worddf[,1], freq = worddf[,2], max.words = 100, scale=c(8,1), 
          random.order = FALSE, random.color = TRUE, colors = customPalette)

Answer 2

What I have observed is that if your input text has any word with frequency greater than or equal to the default value of min freq, ie 3, all other words having freq less than 3 will be ignored.

However, if your input text doesn't have any word with freq >= 3, all words are considered for plotting. So in such cases, always update the min.freq argument and set it to your desired value.