Overview

Martin Westgate & Eliza Grames

2025-02-21

Introduction

Systematic review searches include multiple databases that export results in a variety of formats with overlap in coverage between databases. To streamline the process of importing, assembling, and deduplicating results, synthesisr recognizes bibliographic files exported from databases commonly used for systematic reviews and merges results into a standardized format.

Read and assemble bibliographic files

synthesisr can read any BibTex or RIS formatted bibliographic data files. It detects whether files are more bib-like or ris-like and imports them accordingly. Note that files from some databases may contain non-standard fields or non-standard characters that cause import failure in rare cases; if this happens, we recommend converting the file in open source bibliographic management software such as Zotero.

In the code below, we will demonstrate how to read and assemble bibliographic data files with example datasets included in the synthesisr package. Note that if you are using the code with your own data, you will not need to use system.file() and instead will want to pass a character vector of the path(s) to the file(s) you want to import. For example, if you have saved all your search results in a directory called “search_results”, you may want to use list.files("./search_results/") instead.

# system.file will look for the path to where synthesisr is installed
# by using the example bibliographic data files, you can reproduce the vignette
bibfiles <- list.files(
  system.file("extdata/", package = "synthesisr"),
  full.names = TRUE
)

# we can print the list of bibfiles to confirm what we will import
# in this example, we have bibliographic data exported from Scopus and Zoological Record
print(bibfiles)

# now we can use read_refs to read in our bibliographic data files
# we save them to a data.frame object (because return_df=TRUE) called imported_files
library(synthesisr)
imported_files <- read_refs(
  filename = bibfiles,
  return_df = TRUE)

Deduplicate bibliographic data

Many journals are indexed in multiple databases, so searching across databases will retrieve duplicates. After import, synthesisr can detect duplicates and retain only unique bibliographic records using a variety of methods such as string distance or fuzzy matching records. A good place to start is removing articles that have identical titles, especially since this reduces computational time for more sophisticated deduplication methods.

## first, we will remove articles that have identical titles
## this is a fairly conservative approach, so we will remove them without review
# df <- deduplicate(
#   imported_files,
#   match_by = "title",
#   method = "exact"
# )

In some cases, it may be useful to know which articles were identified as duplicates so they can be manually reviewed or so that information from two records can be merged. Using our partially-deduplicated dataset, we check a few titles and use string distance methods to find additional duplicate articles in the code below and then remove them by extracting unique references. Although here we only use one secondary deduplication method (string distance), we could look for additional duplicates based on fuzzy matching abstracts, for example.

NOTE: the examples below don’t match now; need updating

# there are still some duplicate articles that were not removed
# for example, the titles for articles 91 and 114 appear identical
## df$title[c(91,114)]
# the dash-like symbol in title 91, however, is a special character not punctuation
# so it was not classified as identical

# similarly, there is a missing space in the title for article 96
## df$title[c(21,96)]

# and an extra space in title 47
## df$title[c(47, 101)]

# # in this example, we will use string distance to identify likely duplicates
# duplicates_string <- find_duplicates(
#   df$title,
#   method = "string_osa",
#   to_lower = TRUE,
#   rm_punctuation = TRUE,
#   threshold = 7
# )

# we can extract the line numbers from the dataset that are likely duplicated
# this lets us manually review those titles to confirm they are duplicates

# manual_checks <- review_duplicates(df$title, duplicates_string)

# # the titles under match #99 are not duplicates, so we need to keep them both
# # we can use the override_duplicates function to manually mark them as unique
# new_duplicates <- synthesisr::override_duplicates(duplicates_string, 99)
# 
# # now we can extract unique references from our dataset
# # we need to pass it the dataset (df) and the matching articles (new_duplicates)
# results <- extract_unique_references(df, new_duplicates)

Write bibliographic files

To facilitate exporting results to other platforms after assembly and deduplication, synthesisr can write bibliographic data to .ris or .bib files. Optionally, write_refs() can write directly to a text file stored locally.

# # synthesisr can write the full dataset to a bibliographic file
# # but in this example, we will just write the first citation
# # we also want it to be a nice clean bibliographic file, so we remove NA data
# # this makes it easier to view the output when working with a single article
# citation <- df[1, !is.na(df[1,])]
# 
# format_citation(citation)
# 
# write_refs(citation,
#   format = "bib",
#   file = FALSE
# )