revtools (version 0.4.1)

screen_topics: Shiny app for screening bibliographies using topic models

Description

Screening is usually achieved by manually sorting titles or abstracts one at a time. screen_topics offers an alternative by allowing the user to group data by any column in the input dataset, and running a topic model on the resulting data. This allows a great deal of flexibility to locate patterns in journals, years, or authors, rather than just articles. Data points can be selected or excluded individually, or by topic.

Usage

screen_topics(x, remove_words, max_file_size)

Arguments

x

An (optional) object of class data.frame or bibliography to open in the browser. If empty, the app will launch with no data. Data can be added within the app via the 'import' button.

remove_words

Optional vector of words to be removed from consideration by the Topic Model. If none are given, screen_topics will use revwords. Note that this vector will be converted to lower case before processing, so the algorithm is not case sensitive.

max_file_size

Optional argument to set the maximum file size (in MB) that the app will accept.

Value

This function launches a Shiny app in the users' default browser.

Details

The display space is divided into three parts. From left to right, these are the sidebar; the plot window; and the selection panel.

The sidebar shows a series of drop-down menus that can be used to customize or recalculate the central plot. It can be hidden when not in use. Note that the default settings for LDA (5 topics, 10,000 iterations) prioritize speed over reliability - higher numbers of iterations will give more reliable results.

The plot window shows an ordination of article weights calculated using LDA, with articles colored by their highest-weighted topic. Hovering over a point shows the title and abstract below the plot; clicking allows selection or deselection of that article (and optionally displays co-authorship data). Selecting a region of the plot and clicking zooms on the selected region; double-clicking without selecting a region returns the plot to its full extent.

The selection panel gives information on progress in selecting/deselecting articles. It also contains windows for displaying topic-level infromation and article abstracts. All boxes in this panel can be minimized when not required.

Ordinations are calculated using LDA (library "topicmodels") and are displayed using shiny and plotly.

When you have finished viewing/screening, you can export information to a .csv or .rda file (saved to the working directory) using the 'Save' tab.

Note that "start_review_window" is the earlier form of this function; this has been deprecated and will be removed from future versions of revtools.

See Also

screen_titles or screen_abstracts for manual screening; screen_topics_progress-class for saving and restoring progress in screen_topics.

Examples

# NOT RUN {
# to run the app and upload data interactively
# }
# NOT RUN {
screen_topics()
# }
# NOT RUN {
# or to specify data from the workspace
file_location <- system.file(
  "extdata",
  "avian_ecology_bibliography.ris",
  package = "revtools")
x <- read_bibliography(file_location)
# to run the app using these data:
# }
# NOT RUN {
screen_topics(x)
# }
# NOT RUN {
# or to run the app & save results to the workspace:
# }
# NOT RUN {
result <- screen_topics(x)
# }