AIM review tool: artificial intelligence for smarter systematic review screening
AIM review tool: artificial intelligence for smarter systematic review screening
https://www.nature.com/articles/s44387-026-00080-8
Publish Date: 2026-02-21 07:55:00
Source Domain: www.nature.com
- AIM Review Application: A client-side web application developed using Google’s Firebase platform, designed to operate without server-side connections and run locally in the user’s browser.
- Modular Structure: Comprises three independent components: labeling, agreement assessment, and ML prediction applications. Users can parallelly run these modules for efficient workflows.
- Text Vectorization Methods: Incorporates various methods to convert titles and abstracts into numerical representations suitable for ML classification, including TF-IDF, LSA, universal sentence encoder, Doc2Vec, and pretrained sentence transformers.
- Classification Models: Offers a range of classifiers including logistic regression, support vector machines, decision tree classifier, multi-layer perceptron, and sequential neural networks for predicting publication relevance.
- Active Learning Strategies: Implements active learning techniques that allow user-supervised relevance ranking of publications through classifier models trained on screened entries to prioritize review efforts.
- Nested Cross-Validated Supervision: Utilizes nested cross-validation to optimize model hyperparameters for robust performance, including methodologies inspired by NeuroMiner for reliable model evaluation.
- Ensemble Strategies: Supports both stacked generalization and modality fusion in combining vectorization modalities to enhance prediction robustness.
- Case Studies Evaluation: Demonstrates the application’s effectiveness across six systematic review case studies, testing various fields and search sizes to ensure generalizability and fair comparisons of vectorization and classification methods.
