ReCiter: An open source, identity-driven, authorship prediction algorithm optimized for academic institutions.

Area of Science:

• Bibliometrics
• Information Science
• Academic Publishing

Background:

• Academic institutions face challenges maintaining accurate publication lists for numerous scholars.
• Manual curation is time-consuming and scholars often lack the capacity for consistent updates.
• Existing author disambiguation methods using clustering alone do not meet institutional needs for large-scale article mapping.

Purpose of the Study:

• To introduce ReCiter, an identity-driven authorship prediction application for academic institutions.
• To evaluate ReCiter's effectiveness in programmatically mapping articles to scholars en masse.
• To optimize ReCiter's scoring algorithm using machine learning for enhanced curation efficiency.

Main Methods:

• ReCiter utilizes institutionally-maintained identity data (e.g., department, degree year) to predict authorship.
• It computes article scores based on identity data, cluster acceptance/rejection history, and average cluster scores.
• A support vector machine analysis was employed to optimize the scoring algorithm.

Main Results:

• ReCiter achieved 98% accuracy in predicting publications for a diverse test group of 500 scholars.
• The application provides scoring and qualitative evidence to support suggested article attributions.
• Optimization using support vector machines improved the algorithm's performance.

Conclusions:

• ReCiter offers a scalable and accurate solution for academic institutions to manage scholar publication lists.
• Identity-driven authorship prediction, as implemented in ReCiter, is a viable alternative to traditional methods.
• The system's confidence scoring aids curators in efficient and accurate feedback processes.