Analyzing Social Media Data to Understand Consumer Information Needs on Dietary Supplements

Rubina F. Rizvi, Yefeng Wang, Thao Nguyen, Jake Vasilakes, Jiang Bian, Zhe He, Rui Zhang

TL;DR

This paper uses topic modeling on Yahoo! Answers data to uncover consumer information needs regarding dietary supplements (DS). The findings reveal that consumers are most interested in the uses and adverse effects of DS, product-related information, and how DS fit into a healthy lifestyle, providing a roadmap for creating more targeted and useful DS resources.

What Problem Does This Paper Address?

Despite the widespread use of dietary supplements (DS) in the US, there's a lack of comprehensive, reliable, and easily understandable online resources that cater to consumers' specific information needs. Existing resources are often basic or require subscriptions. Consumers often self-initiate DS use, making it crucial to understand their information-seeking behavior, especially as they turn to the internet for health information. This paper aims to bridge this gap by analyzing the questions consumers ask on social media to identify their key concerns and information gaps related to DS.

Methods

The researchers employed Correlation Explanation (CorEx), an unsupervised topic modeling algorithm, to analyze a large dataset of questions from Yahoo! Answers. CorEx was chosen over Latent Dirichlet Allocation (LDA) because it was found to discover more coherent topics by maximizing the mutual information between words and topics.

The process involved:

  1. Data Collection and Pre-processing: Extracting questions related to "Alternative Medicine" from Yahoo! Answers, filtering for those mentioning specific dietary supplement ingredients. This involved cleaning the data by removing special characters, lowercasing text, tokenizing, and removing stop words and infrequent terms.
  2. Topic Modeling: Applying CorEx to identify latent topics within the pre-processed questions. A model with 200 topics was selected for its coherence.
  3. Topic Labeling and Merging: Domain experts manually reviewed the keywords of the generated topics, assigning them to 38 health-related categories, which were then further grouped into 12 higher-level categories (e.g., "uses and symptoms," "product-related").
  4. Topic Evaluation: Assessing the accuracy of the topic modeling by manually reviewing a subset of questions and determining how well they aligned with the generated topics. This was reported as a percentage accuracy.
  5. Ingredient Analysis: Extracting and analyzing the distribution of specific DS ingredients associated with the identified topics.

Datasets

The primary dataset used was Yahoo! Answers.

  • Initial Collection: 2,820,179 questions from the "Health" category.
  • Filtered Collection: 112,090 questions from the "Alternative Medicine" sub-category.
  • Final Corpus for Analysis: 16,095 unique questions related to 438 unique DS ingredients, resulting in a corpus of 213,790 tokens and 7,164 unique words after pre-processing.

No new benchmarks were introduced; the study utilized existing data from a public Q&A platform.

Key Insights and Ideas

  • Social Media as a Rich Information Source: The study demonstrates that user-generated content on platforms like Yahoo! Answers can provide valuable, unfiltered insights into consumer information needs, complementing traditional research methods.
  • Consumer Focus on Practicality: The most dominant topic groups revolve around the practical aspects of DS use: what they do (uses), what can go wrong (adverse effects), how to use them (product-related), and how they fit into a broader health context (healthy lifestyle).
  • Specificity Matters: The detailed analysis of ingredients within topics highlights specific consumer interests, such as "Honey" for respiratory issues and "Melatonin" for sleep disorders, suggesting areas where targeted information could be highly beneficial.
  • Validation of Topic Modeling: The high accuracy achieved in aligning questions with identified topics validates CorEx as a robust method for extracting meaningful information from large, unstructured social media datasets in health research.

Key Results

  • High Accuracy: The topic modeling approach achieved high accuracy (90-100%) in correctly identifying questions that align with the generated topics, with exceptions for "sleep" (80%) and "frequency/time" (70%) categories.
  • Dominant Topic Groups: The most prominent topic groups were "Use and adverse effects" (accounting for 50 out of 200 topics), "Product-related," and "Healthy life style."
  • Specific Areas of Interest: Within the "Use and adverse effects" group, consumers frequently asked about gastrointestinal issues, psychiatric concerns (anxiety, depression), and skin/subcutaneous tissue issues.
  • Ingredient Insights: "Honey" was a frequently mentioned ingredient in respiratory disorder topics (47.48% of questions), and "Melatonin" was prominent in sleep disorder topics (64.44% of questions).

Limitations and Future Work

Limitations: * Sub-category Bias: The analysis was limited to questions from the "Alternative Medicine" sub-category, potentially missing DS-related queries in other health sections (e.g., mental health, general healthcare). * Ingredient Synonym Handling: The study only used preferred DS ingredient names and did not account for synonyms or common names, which might have excluded relevant questions. * Inherent Topic Modeling Limitations: Topic models can sometimes generate incoherent topics due to statistical word distribution, and the interpretation relies on expert judgment.

Future Work: * The knowledge gained can be used to develop more comprehensive and structured DS resources for consumers. * Further exploration of specific ingredient-topic associations could lead to more targeted information delivery. * Expanding the analysis to include other social media platforms or different sub-categories could provide a more holistic view of consumer information needs.

Summary updated

Abstract

Despite the high consumption of dietary supplements (DS), few reliable, relevant, and comprehensive online resources could satisfy information seekers. This research study aims to understand consumer information needs on DS using topic modeling, and to evaluate accuracy in correctly identifying topics from social media. We retrieved 16,095 unique questions posted on Yahoo! Answers relating to 438 unique DS ingredients mentioned in sub-section, “Alternative medicine” under the section, “Health” . We implemented an unsupervised topic modeling method, Correlation Explanation (CorEx) to unveil the various topics in which consumers are most interested. We manually reviewed the keywords of all the 200 topics generated by CorEx and assigned them to 38 health-related categories, corresponding to 12 higher-level groups. We found high accuracy (90–100%) in identifying questions that correctly align with the selected topics. The results could guide us to generate a more comprehensive and structured DS resource based on consumers’ information needs.
BibTeX
@inproceedings{rizvi2019analyzing,
  author = {Rubina F. Rizvi and Yefeng Wang and Thao Nguyen and Jake Vasilakes and Jiang Bian and Zhe He and Rui Zhang},
  title = {{Analyzing Social Media Data to Understand Consumer Information Needs on Dietary Supplements}},
  booktitle = {MedInfo},
  year = {2019},
  doi = {10.3233/shti190236},
}
Figure