I recently updated this website, and I then decided to include a wordcloud on my start page. This wordcloud represents the titles and abstracts of all papers I’ve saved in my Mendeley library. Here I will now describe how I made this.

First of all, you need to localize the Mendeley SQLite database. I use a Mac, so I can find it in $HOME/Library/Application\ Support/Mendeley\ Desktop/<email>@www.mendeley.com.sqlite. Just replace <email> with the email address you use for Mendeley. If you can’t find the database, take a look at this Mendeley support page. In R, we’ll then extract the relevant columns from the database.

library(RSQLite)

con <- dbConnect(RSQLite::SQLite(), "~/Library/Application Support/Mendeley Desktop/<email>@www.mendeley.com.sqlite")
res <- dbSendQuery(con, "SELECT title, abstract FROM Documents")
dbdata <- dbFetch(res)

Now we have a data frame with two columns containing the title and abstract for every document in the library. We could create a wordcloud directly from this data, but it would probably not look very good since it would contain variants of the same words, i.e. gene and genes or analyzed and analysis. What we want to do is to find the common denominator among these words, and this process is called stemming. I will do this in R using the tm (text mining) package. tm has a data structure called Corpus, but in this simple context, I find it easier to just work with character vectors where each element is a word.

words <- unlist(strsplit(as.character(dbdata), " "))

To make sure that all words look the same, we transform them to lower case and remove punctuation and numbers, and finally stem them.

library(tm)
words <- tolower(words)
words <- removePunctuation(words)
words <- removeNumbers(words)
words.stem <- stemDocument(words)

If we look at these words now, many of them does not make sense. For example, the words “promote”, “promoter”, “promotion” and “promoted” would be all stemmed to the form “promot”, which is not a word (as far as I know). To avoid this gibberish, we can use the stemCompletion function. What it does is that it looks through the stemmed words and transform them to the most common original variant of the stemmed word. This is the default behavior, and you can look at the documentation for stemCompletion for other options.

words.stem <- stemCompletion(words.stem, dictionary = words)

Now we have a character vector of words that should make sense. If we would want to create the wordcloud directly in R, it is possible. However, then you would want to remove common words such as “the”, “of” and “and”. This can be done with the removeWords function together with stopwords that contains a number of stop words for a number of languages. The wordcloud package can then be used to create the actual wordcloud.

In my case I didn’t create the wordcloud in R, but used the online service Wordle. Simply save the words to a text file and just paste it into their service. You can then choose among different color schemes and layouts to produce a wordcloud that you like.