Abstract:
|
Meta-analysis combines summary data from multiple studies to draw a global conclusion. Identifying articles to extract data for meta-analysis is often burdensome and can be a severe bottleneck for investigators. First, investigators develop a study protocol that contains key words that are used to search for relevant articles in search engines, such as PubMed. After the candidate set of articles is found, a more in-depth screening follows, where humans read titles and/or abstracts. If the search is broad, investigators may need to read through hundreds or thousands of titles and abstracts. Articles that pass the title and abstract screen are then read to extract numerical information for the meta-analysis. I discuss text processing methods and statistical and machine learning models used to screen titles and abstracts using natural language. These methods and models aim to reduce the burden of identifying articles for meta-analysis.
|