Abstrackr vs RobotReviewer

Abstrackr is a free web tool for screening citations in systematic reviews. It supports solo work, blinded team screening, pilot rounds, and conflict resolution. After about 100 labels, machine learning ranks remaining citations by predicted relevance. A live PRISMA flow table records screening thresholds, resolved conflicts, exclusion reasons by tag, and analysis of the model’s stop point. Citations can be imported from PubMed, Scopus, or Embase, or uploaded as RIS and CSV files. The service deduplicates records by DOI and PMID and can merge remaining near-duplicates. Results export as CSV or RIS, with a full JSONL download for DuckDB queries. Blinding starts enabled, and changes are logged with user, reason, timestamp, and project snapshots. Member and Leader roles have no stated team-size limit, while reminders arrive 7, 4, and 1 day before deadlines. No installation is needed, and the site directs methodology questions to its community forum.

RobotReviewer analyzes clinical trial reports by annotating PDFs with PICO characteristics and likely bias assessments using the Cochrane Risk of Bias tool. Uploading PDFs on the website extracts key data and presents a summary table. Its REST API can process groups of clinical trial reports to synthesize results across trials. The project is open source under GNU GPL version 3.0 and is free to use. It can run locally as Docker services, with a web app and API server exposed on local ports. The self-hosted processing setup uses Grobid, RabbitMQ, Celery, and SciBERT through bert-as-a-service. Docker instructions cover GPU use, and the worker can also run CPU-only. Manual setup calls for Python 3.6, TensorFlow 1.12.0, and the associated processing services. Installation is documented as tested on Ubuntu and Mac OS; Windows works with substantial installation difficulty. Maintainers ask users to cite the 2017 ACL paper and invite questions by email, including requests for help integrating RobotReviewer into other software.