Using AI to Classify Food Stores
Author Information
Author(s): Etheredge Analee J., Hosmer Samuel, Crossa Aldo, Suss Rachel, Torrey Mark
Primary Institution: Center for Population Health Data Science, NYC Department of Health and Mental Hygiene
Hypothesis
Can large language models effectively categorize food stores based on their names?
Conclusion
The study demonstrated that a large language model can efficiently categorize food stores, significantly improving data quality and operational efficiency.
Supporting Evidence
- The model achieved good performance with F1 scores above 0.7.
- Efficiency gains were noted compared to manual classification methods.
- Known temporospatial patterns in food retail were replicated using the model.
Takeaway
This study shows that computers can help figure out what type of food stores are just by looking at their names, making it easier for health departments to understand food options in the city.
Methodology
The study used a transformer model called BERT to classify food stores based on their names, trained on a dataset of labeled food retail locations in NYC.
Potential Biases
Subjective assessments by surveyors in the training data could introduce bias in classification.
Limitations
Grocery stores were the most challenging to classify, and noise in administrative datasets may affect the accuracy of the model.
Participant Demographics
The study focused on food retail locations in New York City.
Statistical Information
P-Value
0.710
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website