AI and Machine Learning for Evidence Syntheses: What are artificial intelligence and machine learning?
Created by Health Science Librarians
Overview
Welcome to the UNC Health Sciences Library's (HSL) guide to automation and machine learning (ML) for research. This guide helps researchers find, use, and understand how artificial intelligence (AI) tools can be used for research, particularly systematic reviews and other evidence syntheses.
This guide details the differences between predictive and generative AI and how these approaches can be used to complete systematic reviews and other evidence syntheses more efficiently. Several free and commercially available AI tools that may be of interest to enhance the research process are also highlighted in this guide.
Predictive AI for Systematic Reviews
What are Predictive AI Tools?
Regarding evidence synthesis projects, predictive AI tools use machine learning algorithms to analyze and classify research papers, helping researchers streamline the screening process for systematic reviews and other evidence syntheses. These tools can:
- Prioritize relevant articles for review
- Identify patterns in research methodologies
- Reduce screening time
- Maintain systematic review quality while increasing efficiency
Generative AI and Other Tools for Research
We also highlight and compare select generative AI tools for research.
- Generative AI Tools - includes a comparison of generative AI tools, including benefits, limitations, best use, and more.
- AI for Systematic Reviews - includes tools for citation screening, citation searching, data extraction, and quality assessment.
- AI for Research - includes tools for citation searching, citation analysis tools, research planning assistants, and visualization tools.
-
AI for Academic Writing - includes tools to enhance writing, like grammar and style checkers, formatting software, citation managers, and more.
| Term | Definition |
|---|---|
| Artificial Intelligence (AI) | Technology that enables computers to simulate human intelligence and perform tasks that typically require human cognition, such as learning, reasoning, and problem-solving. |
| Active Machine Learning | An iterative process where a machine learning model learns to predict outcomes while humans label data. |
| Deep Learning | A type of machine learning that mimics how humans learn using artificial neural networks, often using complex layers of algorithms. |
| Generative Artificial Intelligence (GenAI) | A type of AI that creates, or generates, new content, such as text or images, based on patterns learned from existing data. |
| Information Hallucination (confabulation) | A phenomenon where an AI model generates content that appears plausible but is actually false or misleading, often based on patterns learned during training rather than verified facts. For example, an AI might create a citation for a study that sounds credible but does not actually exist. |
| Large Language Models (LLMs) | Advanced AI systems that use deep learning techniques to understand and generate human language. Highly skilled in text-based tasks like generation, translation, and summarization. |
| Machine Learning (ML) | A subset of AI where systems learn and improve from experience without being explicitly programmed, using algorithms and statistical models. |
| Machine Learning: Unsupervised | Technique where models identify patterns and group data based on similarities without using training data. Also referred to as clustering. |
| Machine Learning: Semi-Supervised | A type of machine learning that uses some training data to classify unlabeled data. Also referred to as supervised clustering. |
| Machine Learning: Supervised | A method that learns from human-labeled training data to make predictions about the labels for the rest of the dataset. |
| Model Validation | Checking that a machine learning model functions the way it should. There are a variety of different ways to validate models. |
| Natural Language Processing (NLP) | A branch of AI that focuses on the interaction between computers and human language, enabling machines to understand, interpret, and generate text in a meaningful way. |
| Predictive Artificial Intelligence | AI that analyzes historical data to make predictions about future events or outcomes. |
| Testing | The process of evaluating a trained model's accuracy and performance. |
| Text Analytics | The process of analyzing unstructured text data to extract meaningful patterns, insights, and information using computational methods. |
| Training Data | Data that has been labeled or classified by humans, which is used to teach a supervised machine learning model patterns and relationships so it can learn and make predictions. |
| Training Data: Seed Studies |
A type of training data used to help predict outcomes in semi-supervised machine learning. With literature based research, seed studies would be a subset identified as relevant from a randomly screened set. |
Partnering with UNC HSL
HSL librarians have expertise in applying machine learning to search results to minimize the amount of citations that must be screened manually for evidence syntheses.
Available Services Offered by UNC Librarians
- Unsupervised Machine Learning: Clusters citations into groups based on common words. Groups can be used to refine the search strategy.
- Semi-Supervised Machine Learning: Groups citations based on similarity to known relevant citations. Groups with more known relevant citations can be prioritized for screening.
- Supervised Machine Learning: Learns from training data to prioritize citations that are relevant to the project and provides a cut off on when to stop screening with a precision value.
Partner with UNC Librarians
Our expert librarians can help you:
- Select appropriate AI tools for your research needs
- Set up and configure tools for optimal results
- Develop effective search strategies
- Ensure methodological rigor
Contact Us
Current UNC and UNC Health affiliates can contact the UNC Health Sciences Library for more information at https://asklib.hsl.unc.edu/.