Funny item co-occurrences in 3.2M Instacart orders
13 hours ago
- #data analysis
- #grocery shopping
- #humor
- The Instacart dataset from 3 million orders was analyzed to find funny item co-occurrences.
- Initial analysis using product-level data yielded uninteresting results due to too many unique combinations.
- Using 'lift' (actual frequency vs. expected frequency) improved ranking but still lacked humor.
- Products were reclassified into 1,000 GPC bricks for better granularity, reducing complexity by 50x.
- A humor score (0-1) was assigned to each brick based on taboo or comedic potential.
- Funniest rare combos emerged when combining humor scores with lift, such as 'Garlic and Diarrhoea Remedies' or 'Kale and Enemas/Douches'.
- Small carts (2-3 items) also produced humorous pairs like 'Oreo Cookies and Personal Lubricant'.