Examining Analytic Practices in Latent Dirichlet Allocation Within Psychological Science: Scoping Review

Area of Science:

• Psychological Science
• Computational Social Science
• Natural Language Processing

Background:

• Topic modeling, including Latent Dirichlet Allocation (LDA), is a key method for analyzing large text datasets in psychology.
• However, LDA outputs can be sensitive to analytical choices and may not suit all data types, such as short texts.
• Existing guides often overlook the complex analytical practices specific to LDA.

Purpose of the Study:

• To conduct a scoping review of methodological approaches in psychological research utilizing LDA.
• To examine data selection, preprocessing, and analysis steps in LDA applications within psychology.
• To identify areas for improvement in reporting standards and best practices for LDA in psychological science.

Main Methods:

• A systematic search of 4 psychology and health databases was performed.
• Studies employing LDA for analyzing text related to psychological constructs were included.
• A data charting process based on standard analytical steps (selection, preprocessing, analysis) was used.

Main Results:

• 68 studies using LDA in psychology were identified, with many sourcing data from social media.
• Reporting on preprocessing and data analysis steps was often insufficient for reproducibility.
• A debate exists regarding the necessity of specific preprocessing and data analysis techniques in LDA.

Conclusions:

• The use of LDA in psychological science is expanding.
• There is a critical need for enhanced analytical reporting standards and evidence-based best practices.
• An LDA Preferred Reporting Checklist was developed to promote consistent documentation and reproducible research outcomes.