Abstract
Despite the widespread use of dietary supplements, a significant gap exists between the technical information provided by official health agencies and the practical, personalized questions consumers ask. This study employs Natural Language Processing (NLP) and the Correlation Explanation (CorEx) topic modeling algorithm to analyze 16,095 unique questions from Yahoo! Answers. By masking specific ingredient names, the researchers identified 12 high-level information needs, such as "Uses & Adverse Effects" and "Product Dosage," with high alignment accuracy (90-100% for most categories). These findings provide a roadmap for creating more intuitive, consumer-focused health resources.
Introduction
Over half of U.S. adults consume dietary supplements (DS), yet official resources often fail to address their specific concerns. Consumers frequently turn to social media and Q&A platforms for health advice, generating vast amounts of unstructured data. This research aims to "listen" to these concerns at scale, identifying the latent structure of consumer questions to guide the development of better health information systems.
Methodology
The researchers developed a multi-step pipeline to process raw social media text. First, questions were extracted from Yahoo! Answers and matched against the iDISK knowledge base. A critical step involved "Ingredient Masking," where specific supplement names were removed to ensure the algorithm grouped questions by the type of information requested rather than just the supplement name.
The study utilized the Correlation Explanation (CorEx) algorithm, an information-theoretic approach that maximizes mutual information between words and topics. This was found to be more effective than traditional methods like LDA for messy social media data. Finally, 200 generated topics were mapped to 38 categories and condensed into 12 high-level groups using the MedDRA hierarchy for standardization.
Results and Key Insights
The analysis revealed that consumer needs are remarkably consistent across different supplements. The "Uses and Adverse Effects" group was the most prominent, accounting for 25% of all identified topics. This indicates that safety and efficacy are the primary drivers of consumer inquiries on social media.
Other significant findings include:
- High Precision: Manual evaluation confirmed that the model correctly aligned questions with topics with 90-100% accuracy for the majority of categories.
- Ingredient Anchors: Certain supplements are strongly associated with specific health concerns in the public consciousness, such as Melatonin for sleep (mentioned in nearly 65% of sleep-related questions) and Honey for respiratory issues (nearly 48%).
- The "Big Three" Needs: The vast majority of questions fell into "Uses & Adverse Effects," "Product-Related" (dosage/preparation), and "Healthy Lifestyle" (weight control/exercise).
Conclusion
By mapping social media "slang" to professional medical terminologies, this study bridges the gap between consumer language and clinical data. While limited by selection bias (focusing only on the "Alternative Medicine" category), the results suggest that future health resources should be structured around the intuitive ways consumers think and ask questions, potentially using these identified topics as a framework for more accessible FAQ systems.
Summary updated