Text Mining Approaches for Postmarket Food Safety Surveillance Using Online Media.

Area of Science:

• Food safety
• Computational linguistics
• Consumer product monitoring

Background:

• Food contamination and poisoning present global consumer risks.
• Online media increasingly discusses consumer experiences with food safety.
• Traditional methods for monitoring food safety hazards are often slow and reactive.

Purpose of the Study:

• To develop and evaluate a text mining approach for rapidly screening online media for food safety hazards.
• To compare the effectiveness of supervised machine learning against traditional sentiment analysis for identifying hazardous food products.
• To establish a postmarket decision support system for real-time food safety risk assessment.

Main Methods:

• Compiled a large dataset of labeled consumer posts from two major websites.
• Utilized text mining and supervised machine learning to identify unique linguistic markers of hazardous food interactions.
• Compared the developed methods against traditional sentiment-based text mining using over 4 million online reviews.

Main Results:

• The text mining and machine learning approach achieved 77-90% accuracy in identifying high-risk reviews.
• Traditional sentiment analysis showed significantly lower accuracy, ranging from 11-26%.
• A panel of food safety experts validated the model's ability to flag products with substantially higher risk.

Conclusions:

• Text mining and supervised machine learning offer a practical and cost-effective solution for real-time food safety monitoring.
• The developed system can effectively profile food items and assess product-level risks.
• This approach enables the creation of a robust postmarket surveillance system to enhance consumer protection.