Boosting Screening Efficiency in Research Synthesis with Text Mining and Machine Learning

• In Event: AERA Poster Session Wednesday 3:45pm
  In Poster Session: Systematic Review and Meta-Analysis: Methods, Contexts, and Impacts

Abstract

This study proposes a two-stage unsupervised method integrating Latent Dirichlet Allocation (LDA) and K-means++ clustering to improve the efficiency of abstract screening in systematic reviews. By identifying thematically relevant clusters for prioritized review, the method significantly reduces workload while maintaining high recall. Performance is evaluated with a systematic review and a meta-analysis. Results show that the proposed method retains over 90% of human-included studies while reducing the abstract screening pool by over 40%. Compared to GPT-based and LDA-based automated methods, the LDA-K-means method offers a transparent, semi-automated alternative that combines efficiency with interpretability. This work contributes a practical, scalable solution for enhancing research synthesis workflows.

Authors

• Xiyu Wang, Purdue University

• Yukiko Maeda, Purdue University