Skip to content

Latest commit

 

History

History
72 lines (57 loc) · 3.43 KB

README.md

File metadata and controls

72 lines (57 loc) · 3.43 KB

📌 Yelp Business Recommendation System

📜 Overview

This project builds an item-based collaborative filtering recommendation system for businesses using the Yelp dataset. The system integrates table lookups, embeddings, and collaborative filtering to provide accurate business recommendations based on user queries.

🎯 Problem Explanation

Users often search for recommended businesses based on specific categories, attributes, and locations. This project aims to:

  • Retrieve business recommendations using structured queries.
  • Enhance recommendation accuracy with text embeddings and collaborative filtering.
  • Visualize recommendations using an interactive geo-mapping system.

📌 Note: Users dataset is very large. Please contact me if you want to use it.

📌 Dataset: Yelp Business Dataset (including business attributes, reviews, users, and ratings).

⚙️ Installation & Dependencies

Requirements

  • Python 3.x
  • Libraries Used:
    • pandas – Data processing
    • numpy – Numerical computations
    • spaCy – Text processing
    • scikit-learn – ML models, cosine similarity
    • folium – Geospatial visualization
    • geopy – Location services

🔍 Project Structure

1️⃣ Business Recommendation by Table Lookup

📌 Objective: Retrieve businesses based on predefined query formats.

✅ Methodology:

  • Analyzed key attributes (categories, city, attributes) to determine recommendation factors.
  • Transformed category text into a structured format (Coffee & Tea → Coffee, Tea).
  • Implemented:
    • getRec(query): Retrieves businesses based on structured queries.
    • geoRec(query, numRec): Generates interactive maps for recommendations.
  • Applied query filtering logic:
    • Best = Businesses with ratings > 4.0.
    • Category-based and location-based filtering.
    • Integrated geospatial mapping using folium and geopy.

2️⃣ Recommender System using Embeddings

📌 Objective: Predict business ratings using word embeddings and improve recommendation accuracy.

✅ Methodology:

  • Text preprocessing using spaCy (stop word removal, tokenization).
  • Generated word embeddings using GloVe (50D vectors).
  • Computed business recommendations based on dot product similarity.
  • Applied a binning strategy to improve rating predictions.
  • Reduced RMSE from high variance to 1.89 after optimizations.

3️⃣ Item-Based Collaborative Recommendation using Embeddings

📌 Objective: Recommend businesses based on category similarity and embeddings.

✅ Methodology:

  • Refined business categories for better retrieval (Coffee & Tea → [Coffee, Tea]).
  • Implemented:
    • filteredData(query): Extracts businesses matching query attributes.
    • computeRecommendation(query): Computes recommendations using cosine similarity.
  • Query parsing logic:
    • Extracts city name, business name, and categories from user input.
    • Filters businesses based on category similarity (must match at least 3 categories).
  • Integrated geo-mapping for visualization.

📊 Results & Findings

✅ Successful integration of table lookups, embeddings, and collaborative filtering.

✅ Improved categorization enhances recommendation accuracy.

✅ Binning techniques significantly refined rating predictions.

✅ Challenges in query sensitivity were mitigated through text preprocessing.