Restaurant-Recommendation-System

Introduction

This project aims to develop a robust Restaurant Recommendation System that serves a dual purpose. For customers, it provides personalized restaurant suggestions based on individual preferences and geographical proximity. For restaurant owners, it offers a data-driven approach to generating targeted advertisements to attract customers effectively, leveraging insights from user reviews.

For Customers:

The system utilizes a sophisticated recommendation engine to curate a list of restaurants tailored to a user's tastes and dining history. By considering factors such as past reviews, preferences, and location data, the system ensures that users receive the most relevant and appealing recommendations.

For Restaurant Owners:

On the business front, restaurant owners benefit from an advanced data mining application that analyzes association rules within customer data. This facilitates the creation of impactful advertisements aimed at drawing in a crowd more likely to appreciate what the restaurant has to offer.

Technical Approach:

At the core of the recommendation engine is a hybrid model that combines the strengths of Collaborative Filtering and Content-Based Filtering. We use algorithms such as KNN, KNNWithMeans, and SVD for Collaborative Filtering and employ BERT embeddings and TF-IDF for Content-Based Filtering. The synergy of these techniques allows for a nuanced understanding of user preferences, leading to highly relevant recommendations.

Project Scope:

The Restaurant Recommendation System is an educational project designed to demonstrate the capabilities of modern recommendation algorithms. It handles a substantial dataset of approximately a million entries sourced from the publicly available Yelp dataset on Kaggle, making it a project that not only educates but also scales to real-world application scenarios.

Our mission with this project is to showcase how data-driven technologies can transform the dining experience for customers and the marketing efforts of restaurant owners.

Table of Contents

• Introduction
• Features
• Project Structure
• Data Preparation
• Models
• Evaluation
• Contributors
• License

Features

For End-Users

• Personalized Recommendations: The system utilizes user past behavior to recommend restaurants that match their taste and preferences.
• Proximity-Based Suggestions: Recommendations are not only tailored to preferences but also to the proximity of the user, ensuring that suggested restaurants are within a reasonable distance.
• Diverse Cuisine Options: By analyzing extensive data sets, the system offers a wide variety of cuisines, catering to different dietary needs and culinary tastes.
• User Review Insights: Incorporates insights from user reviews using advanced NLP techniques, providing a more nuanced recommendation that takes into account the sentiment and context of user reviews.

For Restaurant Owners

• Advertisement Generation Tool: Employs Association Rules Data Mining to help restaurant owners create targeted advertisements that resonate with likely customers.
• Customer Preference Analytics: Provides analytics on customer preferences and dining habits, enabling restaurant owners to tailor their services or menus accordingly.
• Competitive Analysis: Offers insights into competitors and market trends, allowing restaurant owners to position their advertisements more strategically.

Project Structure

The project is organized into several directories, each serving a specific role in the recommendation system:

CollaborativeFiltering

• This directory contains the code and algorithms related to the Collaborative Filtering recommendation technique. It includes implementations of KNN, KNNWithMeans, and SVD algorithms.

ContentBased

• Here, the Content-Based Filtering logic is implemented. It utilizes NLP techniques and machine learning models like BERT and TF-IDF to analyze and match user preferences with restaurant features.

EDA

• Contains Jupyter notebooks and scripts used for Exploratory Data Analysis. This is where the data is explored and visualized to understand patterns and insights.

HybridModel

• This directory integrates the Collaborative and Content-Based Filtering approaches to form a Hybrid Model, leveraging the strengths of both techniques for improved recommendation performance.

Data Preparation

Details

• How the data is prepared for the recommendation system.

1. From Kaggle Combine review.json and business.json
2. Select the most reviews city-Philadelphia.
3. Sample with users who reviewed more than 20 Yelp shops.

• Mention of tf-idf and BERT in the context of data preparation.

Models

• Explanation of the models used in Collaborative Filtering, Content-Based filtering, and the Hybrid model.

1. User-Based Collaborative Filtering: This model recommends items by identifying users with similar preferences and suggesting items those similar users have liked.
2. Item-Based Collaborative Filtering: This model suggests new items based on a user's past preferences, focusing on the similarity between items themselves rather than user similarity.

• Any tuning or optimization that was performed.

The KNNBasic() algorithm serves as a foundational model supporting both user-based and item-based methods. Key parameters of this algorithm include:

• k: Denotes the count of nearest neighbors the model considers in its prediction-making process.

• min_k: Specifies the minimum required count of neighbors for the algorithm to proceed with a recommendation, ensuring a robust and reliable prediction.

Evaluation

The Restaurant Recommendation System's performance is rigorously evaluated using a Hold-Out Test Dataset to ensure the robustness and accuracy of the recommendations provided.

Metrics

For each user-item pair, the system predicts a rating and compares it with the ground truth (GT) rating. The evaluation metrics used are:

• Mean Squared Error (MSE): Measures the average of the squares of the errors, i.e., the average squared difference between the estimated values and the actual value.
• Mean Absolute Error (MAE): Represents the average absolute difference between the predicted ratings and the actual GT ratings.

Results Interpretation

The effectiveness of the recommendation engine is inversely proportional to the RMSE values obtained during testing. Lower RMSE values indicate higher accuracy in the predictions, reflecting a more reliable and trustworthy system from a user's perspective. Our aim is to continually refine the model to minimize RMSE, thereby enhancing the quality of recommendations for the end-users.

Contributors

• Ayush Agarwal
• Kevin Li
• Yumin Zhang
• Grace Xie

License

This project is made available under the MIT License. For more details, see the LICENSE file in the repository.