Food_Recommendations

Recipe Recommendation System 🍲

This is a recipe recommendation system that suggests recipes based on user-provided keywords using TF-IDF vectorization and Word2Vec embeddings. It also evaluates recommendations with Precision@K and Recall@K to measure relevance and recommendation quality.

📜 Project Overview

The model recommends recipes based on descriptive keywords like "spicy vegetarian" or "sweet dessert." This system leverages a combination of TF-IDF vectorization and Word2Vec embeddings to capture context in recipe descriptions, ingredients, and tags. A synonym dictionary further enhances keyword search relevance, capturing alternative expressions for key terms. Evaluation metrics include Precision@K and Recall@K, providing insight into recommendation accuracy.

🚀 Features

• Keyword-based Recipe Recommendation: The model returns recipes that match user-entered keywords, such as "bitter coffee" or "sour lemon."
• Synonym Expansion: Keywords are expanded using a built-in synonym dictionary, e.g., "spicy" includes "hot" and "peppery."
• Evaluation Metrics: Precision@K and Recall@K metrics are included to assess the quality of the recommendations.
• TF-IDF & Word2Vec Embeddings: Combines TF-IDF for initial text vectorization with Word2Vec embeddings for contextualized word representations.

📥 Installation

To use this model locally, ensure you have Python 3.7+ and install the required packages:

pip install pandas numpy scikit-learn gensim

🛠️ Usage

1. Load the model and dataset. The dataset should contain recipe names, tags, ingredients, and descriptions, with a preprocessed combined_text field combining these features.
2. Enter keywords as input, and the model will return the top recipe recommendations and their relevance metrics.

# Example usage
keywords = "sweet dessert"
recommendations = recommend_recipes(keywords, top_k=10)
print("Recommended Recipes:", recommendations)

1. Evaluate the recommendations with Precision@K and Recall@K:

precision, recall = precision_recall_at_k(recommendations, ["sweet", "dessert"], k=10)
print(f"Precision@K: {precision}, Recall@K: {recall}")

📁 Code Structure

• load_and_preprocess_data: Loads and preprocesses the dataset, merging tags, ingredients, and descriptions.
• expand_keywords: Expands search keywords with synonyms for broader and more relevant searches.
• get_embedding: Calculates Word2Vec embeddings for each recipe to capture word context.
• recommend_recipes: Returns recommendations by comparing cosine similarity of keyword embeddings with recipe embeddings.
• precision_recall_at_k: Evaluates Precision@K and Recall@K, measuring the relevance of recommendations.

🔍 Evaluation Metrics

• Precision@K: Measures relevant tags in the top K recommendations.
• Recall@K: Measures how many relevant tags are found in the top K recommendations from all possible relevant tags.

🧩 Example Workflow

1. Run the Model

   keywords = "sour lemon"
   recommendations = recommend_recipes(keywords, top_k=10)
   

2. View Recommendations

   print(recommendations[['name', 'description']])
   

3. Evaluate Recommendations

   precision, recall = precision_recall_at_k(recommendations, ["sour", "lemon"], k=10)
   print(f"Precision@K: {precision}, Recall@K: {recall}")
   

This model provides recipe recommendations based on descriptive keywords and evaluates their relevance using precision and recall.