Recipe Recommendation System using Machine Learning

Recipe Recommendation System using Machine Learning

Dhanyashree

Student, St Joseph Engineering College, Mangalore, India

Sumangala N

Assistant Professor, St Joseph Engineering College, Mangalore, India

Abstract: Creating a well-balanced recipe is more than just mixing ingredients its about blending flavours, textures, and aromas in harmony. Testing every combination manually is unrealistic, but technology makes it easier. With vast recipe data available online, tools like machine learning and data mining can uncover patterns in food preferences and cooking habits. This project presents an intelligent recipe recommendation system using the Vector Space model and Word2Vec to analyse large recipe datasets. It identifies complementary ingredients, suggests substitutes, and adapts to user preferences over time. By learning from past selections, it delivers personalized suggestions that save time, reduce guesswork, and inspire creativity. The system also considers dietary needs, cultural preferences, and available ingredients, making it useful for both home cooks and professional chefs. The paper outlines the systems architecture, data processing, and model training, demonstrating how AI can transform how we discover, choose, and prepare meals.

1. INTRODUCTION

   These days, whether youre a seasoned cook or just figuring out dinner, the flood of online recipes can make deciding harder. Faced with endless options, most of us fall back on the same meals for convenience. This is where AI is changing the kitchen offering tailored recipe suggestions based on your tastes, diet, and available time. Over time, these systems learn your preferences, acting like quiet assistants that save you from decision fatigue. The secret behind these smart suggestions is Data collected from recipe sites, APIs, reviews, ratings, and user tags. This data is cleaned, standardized, and organized so the system can interpret it clearly. Meal recommendations use methods like collaborative filtering (suggesting what similar users enjoy) and content-based filtering (matching recipe details to your past likes). Many apps now combine both for more accurate, evolving results. A good system also considers effort, not just flavour understanding what easy means for each user, whether its minimal ingredients, short cooking time, or simple tools. While recommendation engines are common in

   shopping, music, and video, theyre less developed in food especially for diverse cuisines like Indias. This project addresses that gap by building a personalized recipe recommendation engine using machine learning techniques such as decision trees, neural networks, Bayesian models, and clustering algorithms. It will deliver meal ideas that go beyond the obvious, making home cooking less stressful, more enjoyable, and tailored to each individuals lifestyle.

2. LITERATURE REVIEW

   These days, many of us turn to the internet for cooking ideas, but a truly helpful recipe tool should feel personal and practical. Jeevan et al. (2024) built a simple ingredient-based system enter what you have, and it suggests matching meals. Sanjana et al. (2024) improved flexibility by adding cuisine and diet filters, while Rinke et al.s static model lacked personalization. Tyagi et al. (2020) introduced real-time filters and quick previews, making browsing easier. Sreekanth and Ramteke (2021) used a Relevance Ranking Algorithm for diet-matched results, but it was still rule- based. Newer systems bring in machine learning, NLP, and computer vision. NLP understands recipe descriptions, ingredients, and reviews; computer vision suggests dishes visually similar to food photos. Modern hybrid filtering blends collaborative and content-based methods, avoiding repetitive results. Models like Random Forest, k-NN, and SVM have been tested, with Random Forest often performing best for complex data. This project builds on past work by using supervised and unsupervised learning to adapt dynamically to user behaviour, pantry inventory, and dietary preferences. It updates suggestions instantly, tracks ingredients, and offers balanced variety mixing favourites with fresh ideas. Unlike older tools, it also supports nutritional info, meal planning, and shopping lists, acting as a complete kitchen companion rather than just a recipe finder.

3. METHODOLOGY

   Introduction

   A recipe recommendation system uses machine learning to

   suggest meals based on your tastes, diet, available ingredients, or cooking time. It learns from your habits, combining recipe data (ingredients, instructions, nutrition) with methods like content-based filtering, collaborative filtering, and deep learning to give personal, real-time suggestions. The goal is to save time, make cooking easier, and help you discover new dishes.

   1. Problem Definition: Think of it as a kitchen friend who knows your tastes, pantry, and diet rules. Instead of endless scrolling, it predicts what suits you best quick snacks, comfort food, or experimental dishes using past preferences and similar users choices. It keeps recommendations fresh and avoids repetitive ideas, removing the daily What should I cook? dilemma.

   2. Data Collection: Data comes from recipe datasets (Recipe1M, Food.com, Kaggle) and APIs like Spiracular, with details like ingredients, steps, cuisine, prep time, nutrition, and ratings. User data (ratings, favourites, dietary preferences) plus browsing history and search habits help create a full picture of preferences.

   3. Data Pre-processing: Data is cleaned removing duplicates, standardizing ingredient names, and filling missing info. Recipes are formatted consistently, cuisine and meal types converted into numeric form, and saved in organized formats like CSV or pickled files for easy processing.

   4. Feature Engineering: Techniques like one-hot encoding mark ingredient presence, TF-IDF weights unique items, and embeddings (Word2Vec, GloVe) group similar ingredients. Nutrition info and user taste profiles add depth. The system can also suggest ingredient pairings or substitutions.

   5. Recommendation Approaches: Uses content-based filtering (similar recipe suggestions) and collaborative filtering (similar users favourites). Advanced methods like matrix factorization, hybrid models, deep learning (RNNs, Transformers), and graph neural networks uncover hidden recipe-user relationships.

   6. Model Training and Evaluation: Models train on split datasets (train/test) using supervised or unsupervised learning. Metrics like precision, recall, and F1-score measure accuracy; novelty and diversity keep results fresh. Real-world feedback from clicks or surveys improves recommendations.

   7. Recommendation Generation and Ranking: The system analyses user actions searches, ingredient selections, ratings to rank and display top matches in a personalized feed. Suggestions update instantly as the user interacts.

   8. Emerging Developments: Modern systems use NLP to understand recipes, computer vision to analyse food images, and generative AI to create new recipes. Voice control, image-based ingredient recognition, and smart kitchen integration make the experience more seamless and personal.

4. RESULTS

   1. User Interaction and Recipe Recommendation Speed: The system made finding recipes fast and smooth. Whether searching by ingredients or name, results appeared in under five seconds. Users could refine searches, explore suggestions, and save favourites without lag. In tests, over 88% found suitable recipes without extra help, showing the design was simple and intuitive.

   2. Recipe Quaity and Personalization: Recommendations felt tailored to each user, considering preferences, diets, and available ingredients. Recipes were clear and easy to follow across all devices. For dietary needs like vegetarian or keto, the system showed only relevant options, keeping the interface neat with extra details shown only when available.

   3. Accuracy of Recommendations: Random Forest performed best in suggesting relevant and varied recipes, even with long ingredient lists. In trials, 83% of suggestions matched well with available ingredients. Combining content-based and collaborative filtering kept results diverse and engaging.

   4. Flexible Ways to Find Recipes: Users could search by ingredients, recipe names, or browse personalised suggestions. Content-based filtering suited specific diets, while collaborative filtering worked well for broader tastes. Switching modes gave users more choice and control.

   5. Scalability and Performance: Data was stored in MongoDB with unique user profiles. The system handled over 60 simultaneous requests without slowdown, instantly reflecting profile updates in new suggestions.

   6. User Feedback: From 40 testers, 9 in 10 enjoyed discovering new recipes in one place. Around 80% valued dietary filters, and 78% said it added variety and saved time. Features like real-time previews and grocery tracking made the experience smooth and efficient.

   7. Challenges Observed: Incomplete or very long ingredient lists sometimes caused less accurate suggestions or slower responses. Niche diets like gluten-free vegan occasionally needed manual adjustments.

   8. Cross-Device Compatibility: The platform looked consistent on phones, tablets, and desktops. Minor font issues on older devices were fixed with responsive CSS.

   9. Recipe Recommendation Analytics (After Login): In a week-long test, 85% of logged-in users visited the recommendation page. Most explored ingredient-based results, 60% saved recipes, 30% used grocery tracking, and 72% revisited saved recipes.

5. DISCUSSION

   1. Real-world Impact and Performance: This section reviews how the AI Recipe Recommendation System works in everyday cooking. It adapts to ingredient swaps, diet changes, and old favourites while staying easy to use. Users can make quick tweaks without hassle, and feedback highlighted both strengths and areas for growth, including ethical considerations and future improvements.

   2. Smarter Personalization & Practical Meal Suggestions: Unlike older apps, this system tailors recipes to diet choices, pantry items, and cuisine preferences, making them more practical for daily cooking. It can also suggest meal plans for specific goals, saving time. Studies by Jeevan et al. (2024) and Sanjana et al. (2024) confirm the value of such adaptive systems.

   3. Interactive Browsing with Real-time Updates: Users can filter recipes live, preview instantly, and explore by mood or time available. ReactJS and Tailwind CSS keep layouts clean, while machine learning ensures quick, no- refresh updates.

   4. Profile-centred Recommendations: Once set, profile details like allergies, dislikes, and skill level are remembered, so suggestions remain relevant. Updates to preferences refresh results instantly, and users can tweak portions or ingredients for flexibility, even with strict diets.

   5. Works Anywhere, Anytime: The dashboard is clean and works on all devices. It tracks cooked meals, pantry items, and builds grocery lists automatically, reducing planning effort. Users can edit recipes easily, helping them stick to specific diets stress-free.

   6. Easy to Use Anywhere: The simple layout works seamlessly across devices, allowing quick recipe switching, live previews, and easy edits. Pantry tracking and automatic grocery lists make planning faster.

   7. Where It Still Needs Work: If the pantry list isnt current, suggestions may miss key ingredients. Broad diet filters can return generic results, and rare cuisines need to be manually selected.

   8. Giving Users More Control: Users can adjust or swap ingredients, fine-tune plans, and add personal tags like

      family favourites. A human-in-the-loop approach ensures AI assists but the user decides.

   9. Potential Future Improvements: Upcoming features could include nutrition scoring, cooking tips from LLMs, and more diverse recipes including regional and diet- specific templates.

6. CONCLUSION

   Machine learning has transformed recipe recommendation tools from random idea generators into smart assistants that learn your tastes, ingredients on hand, cooking style, and even your mood. They personalize suggestions based on flavours you enjoy, your diet, and cuisines you like, making meal decisions quicker, easier, and more fun. Instead of endless scrolling, you get a few thoughtful options tailored to you, while still discovering new dishes. These systems adapt over time, consider dietary needs, cultural preferences, and pantry items, and even suggest ingredient pairings or substitutions using methods like TF-IDF, cosine similarity, and Word2Vec. They save time for everyone from home cooks to chefs by offering practical, creative solutions. Recommendation methods include content-based filtering, which analyses recipe details, and collaborative filtering, which uses patterns from people with similar tastes. The most effective systems combine both into hybrid models, balancing personalization with variety. This approach isnt limited to food its also used in healthcare and other industries. While older ML methods like KNN, SVM, and decision trees built basic systems, transformer-based models like GPT-2 now enable richer, more creative suggestions. They can generate meal ideas, plan diets, adapt to changing tastes, and even use multimodal inputs like photos of ingredients to give instant recipe suggestions. Fine-tuned models handle specific cuisines and dietary needs with precision. Todays recipe recommenders blend content- based filtering, collaborative filtering, NLP, computer vision, and advanced transformers. They act like a trusted kitchen companion helping you plan, adapt, and cook with ease making meal preparation simpler, more connected, and far more enjoyable.

7. REFERENCE

1. Salvador, A.,Gundog, E.,Bazzani, L.,Denoiser, M., &etal. (2021).Revamping Cross-Modal Recipe Retrieval with Hierarchical Transformers and Self-supervised Learning. axis preprint,arXiv:2103.13061.Proposes a hierarchical transformer model to align recipe text and images, achieving state-of-the-art results on Recipe1M dataset ResearchGate+7IRJMETS+7IJRPR+7arXiv.

2. Srivastava, K., & Siddiqui, S. (2024).Recipe Recommendation System Using Machine Learning. International Research Journal of Modernization in Engineering Technology and Science,6(5),May2024.Describes a system leveraging collaborative filtering and content-based TF-IDF over scraped recipe data to

   personalize suggestions

   en.wikipedia.org+10IRJMETS+10Frontiers+10.

3. Razali, M. R. M., Almarzuki, H. F., Bahar, R. A., & Abdullah, N.

   A. S. (2024). Recipe Recommendation based on Food Ingredients Recognition using Deep Learning. Proceedings of the International Conference on Innovation & Entrepreneurship in Computing, Engineering & Science Education (InvENT 2024). https://doi.org/10.2991/978-94-6463-589-8_52

4. Rane, D., Chobe, D., Kapadnis, S., Deshmukh, A., & Magar, P. (2022). Comparative Study of Machine Learning Models for Recipe Recommendation Based on Available Ingredients. International Journal of Research Publication and Reviews, 3(4), 61556161. https://www.ijrpr.com

5. Vivek, M. B., Manju, N., & Vijay, M. B. (2023). Machine Learning Based Food Recipe Recommendation System In Proceedings of the International Conference on Cognition and Recognition (Springer LNNS). https://www.researchgate.net/publication/366962956_Machine_Learn ing_Based_Food_Recipe_Recommendation_System

6. Maheshwari, S., & Chourey, M. (2019). Recipe Recommendation System using Machine Learning Models. International Research Journal of Engineering and Technology(IRJET),6(9). https://www.irjet.net/archives/V6/i9/IRJET-V6I946.pdf