The new version of revtools is up on CRAN and GitHub now. There are four major changes in this version.

First, there are three new functions for working with a team. This will help people who want to run systematic reviews as a group, particularly if you want to customize the amount of overlap between reviewers or have different reviewers screen different numbers of articles. There are three new functions for this task:

  • allocate_effort determines what proportion of articles are given to which reviewers
  • distribute_tasks splits a dataset into sub-sections for each reviewer
  • aggregate_tasks re-joins results from different reviewers

Second, there have been several updates to how data is displayed in revtools shiny apps:

  • all have a max_file_size argument allowing the user to upload larger files than previously
  • all apps now accept drag-and-drop of multiple files simultaneously
  • screen_titles and screen_abstracts now allow user-defined sorting (by choosing a column to sort by)
  • accept/reject, navigation buttons, and progress indicators have been relocated to the header bar
  • screen_abstracts now allows navigation by +/- 10 articles

Third, read_bibliography has had some changes to improve functionality and versatility. The biggest change is that this function can now accept a list of file paths and return a single object (by default this is a data.frame). It also will no longer group unknown tags into a single column labelled ‘further_info’; instead it will import these tags as columns without changing their labels. This makes for larger data frames with less interpretable names, but is more true to the nature of the source data.

Finally, version 0.4.0 fixes a bug in the code passed to the tm library when building document-term matrices with make_dtm. Specifically, the section of the code that removed rare terms was non-functional, leading to articles containing rare terms to drift towards the extremes of the main scatterplot in screen_topics. This has now been corrected and supplemented with an option to also exclude very common terms. Consequently, if you re-run your analyses you should see a more even distribution of points with fewer extreme values. If you’re feeling adventurous, you might wish to consider adjusting the upper and lower thresholds for rare/common term exclusion and see what effect it has on the resulting models, as preliminary testing suggests adjusting the defaults can improve fit in some cases.