Health Claim Perception & Consumer Risk: Meal Replacement Case Study

Work in Progress NLP, Consumer Risk Analysis

Consumer video reviews analysis

Project Overview

I analyzed health-related comments on meal replacement products from YouTube videos. Used YouTube API v3 and Google Cloud to scrape 684 comments across 21 videos from 2020–2025.

Around 24% of comments mention health effects, like digestive issues, allergic reactions, or weight/metabolic concerns. This is a significant signal for brands to monitor and act on.

Automated sentiment tools can misclassify risk, so human oversight is critical before making marketing or safety decisions.

Business Insights: Consumer Health & Engagement

⚠ Health Signals

24%

of comments report health effects

Action: Monitor & improve labeling

🚨 Severe Cases

19

Higher-severity mentions

Action: Track & educate consumers

💡 Sensitive Themes

326

Eating disorder mentions

Action: Proactive communication

🌟 Positive Health

11.8%

Positive mentions

Action: Highlight in marketing

🔥 Engagement

28%

High engagement comments

Action: Activate top users

👥 Community Advocates

25

Repeat commenters

Action: Gather early feedback

🤖 AI Monitoring Caution

⚠️

AI misclassification risk

Action: Combine AI with human review

Key Findings

Side Effect Prevalence

  • Health-related mentions: 199
  • Affected comments: 164
  • Share of all comments: ~24%

Distribution by Category

  • Allergic / immune: 78 mentions
  • Mild digestive: 44 mentions
  • Weight & metabolic: 42 mentions
  • Cognitive / neurological: 14 mentions

Higher-Severity Signals

  • Severe digestive: 6 mentions
  • Cardiovascular-related: 5 mentions
  • Energy & mood-related: 8 mentions

Temporal Risk Analysis

  • Anomaly detection: Enabled
  • Spikes detected: None
  • Pattern: Stable baseline

Data reflects consumer-reported experiences. Not medical advice. Use for trend and risk monitoring.

Temporal & Engagement Patterns

Peak Activity Month: 2021-01

Most Active Day: Tuesday

Most Active Hour (UTC): 21:00

Likes per comment are 0.66 correlated with exclamation mark count → more expressive comments get more engagement.

Negative health-related themes get the most engagement → possibly because negative review videos attract like-minded commenters or those seeking validation.

Comment volume over time
01 - Temporal patterns of comment activity
High vs low engagement comparison
02 - Comparison of high vs low engagement comments
Feature correlation matrix
03 - Correlation matrix of key comment features

Current Status

Project is ongoing. Focus now: transcript collection, claim categorization, early risk analysis. Next: iterative modeling and evaluation.

Code & Resources

Full code, including scraping and risk models, is on GitHub.

View Code on GitHub
© 2025 Made by Yuliia Chekhovska. All rights reserved.