Automation of duplicate record detection for systematic reviews: Deduplicator

Area of Science:

• Bibliometrics
• Information Science
• Systematic Review Methodology

Background:

• Duplicate record identification is a critical step in multi-database systematic reviews.
• Manual deduplication is time-consuming and prone to errors.
• Novel automation tools are needed to improve efficiency and accuracy.

Purpose of the Study:

• To introduce and evaluate the efficacy of a new systematic review automation tool, the Deduplicator.
• To compare the Deduplicator's performance against a semi-manual EndNote method for duplicate record removal.
• To assess the performance of different algorithms within the Deduplicator tool.

Main Methods:

• The Deduplicator tool was developed and tested using 10 Cochrane systematic review search results.
• Researchers compared the Deduplicator's 'balanced' algorithm with a semi-manual EndNote method.
• Three Deduplicator algorithms ('balanced', 'focused', 'relaxed') were evaluated on previously deduplicated datasets.

Main Results:

• The Deduplicator reduced deduplication time to 5 min per 1000 records, versus 15 min with EndNote.
• The Deduplicator exhibited a lower error rate (1.8 per 1000 records) compared to EndNote (3.1 per 1000 records).
• The 'balanced' algorithm achieved a high F1 score (0.9647), 'focused' had high accuracy (0.9798) and recall (0.9757), and 'relaxed' showed high precision (0.9896).

Conclusions:

• The Deduplicator tool effectively reduces deduplication time while maintaining or enhancing accuracy in systematic reviews.
• The study highlights the potential of automated tools for streamlining systematic review processes.
• Further comparative research is suggested to position the Deduplicator against a wider range of deduplication methods.