Measuring Psychological Constructs from Social Media Text Using the Word Embedding Projection Approach.

Area of Science:

• Computational Social Science
• Psychological Measurement
• Natural Language Processing

Background:

• Social media text presents opportunities for tracking psychological constructs longitudinally.
• Existing measurement methods require interpretability and stability over time.
• Conceptual Space Theory and semantic projection provide a theoretical basis for new measurement approaches.

Purpose of the Study:

• To develop and evaluate the Word Embedding Projection Approach (WEPA) for psychological measurement using social media text.
• To assess WEPA's stability, interpretability, and predictive validity in a longitudinal setting.
• To demonstrate WEPA's utility in analyzing psychological constructs within a fitness-oriented social media platform.

Main Methods:

• Developed WEPA, representing psychological constructs as semantic axes in a word embedding space using theory-driven anchor words.
• Utilized panel data from a Chinese fitness social media platform (Keep) with over 177,000 users.
• Conducted four studies involving anchor-word validation, human-coded benchmarks, cross-temporal stability checks, and predictive analyses.

Main Results:

• WEPA demonstrated strong agreement with human-coded benchmarks for goal specificity and physiological states.
• WEPA achieved broader text coverage compared to a dictionary-based baseline.
• WEPA scores showed theoretically interpretable predictive associations with subsequent exercise duration, particularly for goal-setting and self-efficacy constructs.

Conclusions:

• WEPA offers a promising, theory-driven approach for psychological measurement from domain-specific social media text.
• The method provides interpretable and stable indicators of expressed psychological salience.
• WEPA enhances the potential of social media data for longitudinal psychological research.