ReCiter

ReCiter predicts scholarly authorship and maintains institutional publication lists by using institutionally-maintained identity data to disambiguate authors and map articles to scholars.

Key Features:

• Identity-Driven Authorship Prediction: Uses identity data points including department names and years of terminal degrees to assess the likelihood that a scholar authored specific publications.
• Comprehensive Data Integration: Integrates up to 12 types of identity data, considers accepted or rejected statuses of other members of an article cluster, and incorporates the cluster average score into predictions.
• Scoring and Evidence Provision: Provides numerical scoring metrics and qualitative evidence for suggested articles to support curator validation.
• SVM-Optimized Scoring Algorithm: Employs support vector machine (SVM) analysis to refine and optimize the scoring algorithm.
• Data Sources and Candidate Scoring: Calculates overall scores for candidate articles drawn from PubMed and optionally Scopus by integrating multiple data points and cluster analysis.
• Architecture and Deployment: Implemented as a Java application with RESTful microservices and uses DynamoDB for data storage.
• Performance Evaluation: In a test of 500 scholars at an academic private medical center, correctly predicted 98% of their publications in PubMed.

Scientific Applications:

• Institutional Publication Maintenance: Automates creation and updating of institutional publication lists by mapping articles to scholars using identity data.
• Author Disambiguation in Bibliographic Databases: Resolves author identity in databases such as PubMed and Scopus using institutionally-maintained identity attributes and cluster analysis.
• Curation Support and Notification: Supplies scored candidate articles and supporting evidence to enable curator validation and notification of scholars about attributed publications.

Methodology:

Uses institutionally-maintained identity data to predict authorship, calculates overall scores for candidate articles from PubMed (and optionally Scopus) by integrating multiple identity data points and cluster analysis, considers accepted/rejected cluster member statuses and cluster average scores, and applies SVM analysis to optimize the scoring algorithm; implemented as a Java application with RESTful microservices and DynamoDB for storage.