Some restaurant customers who contract foodborne illnesses do not contact public health authorities but instead post online reviews to social media. By monitoring social media discourse, health authorities can gather information updates about restaurants’ hygiene deficiencies and thereby identify potential venues for outbreaks of foodborne illness. This study proposes a social media analytics framework to analyze the associations among negative hygiene aspects mentioned in customers’ reviews and use those associations to predict restaurants’ food safety.
This study analyzes customer reviews of restaurants and identifies the co-occurrence patterns of hygiene-related keywords. To assess the extent to which the word co-occurrences are effective in preventing foodborne illnesses, classification models were constructed to use those co-occurrences as inputs to predict restaurants’ food safety risk.
This study obtains 20 association rules that reveal the co-occurrences of hygiene-related keywords. Using those co-occurrences as inputs, our best-performing model can detect 87.58% of high-risk restaurants.
When monitoring social media, health authorities can focus on a group of keywords and deploy our model to identify restaurants that are likely to contribute to foodborne illnesses.
Through the lens of signaling theory, this study is a pioneering work to reduce the dimensionality of social media data to a few meaningful hygiene-related keywords, filtering out irrelevant signals that disturb the signaling process. Social media data, after being processed by appropriate machine learning algorithms, become credible signals for risk prediction.
