Case study
Recipe recommendation system
Finding healthy recipes that fit dietary preferences, nutrition goals and the ingredients on hand is hard. This recommender weighs nutrition profiles, taste preferences and dietary restrictions together.
- Recipes
- 20K+
- Features
- 680
- Recommendation tiers
- 3
Approach
- Curated and cleaned 20,000+ recipes with nutrition, ingredients, ratings and preparation details.
- Engineered nutrition features: normalized macros, calorie density ratios and nutrient balance scores.
- Built three tiers: content-based filtering with cosine similarity on nutrition vectors, ingredient matching with personalized weights, and a hybrid score that combines both.
- Applied dietary filters (vegetarian, vegan, gluten-free, keto and more) before scoring.
Results
- Recommendations sit close to the target nutrition profile by cosine similarity, the score the system optimizes.
- Because filters run before scoring, restricted recipes never reach the recommendations.
- A statistical test (p < 0.05) showed that recommended recipes differ from random picks on key nutrition metrics.