https://github.com/pngo1997/learning-to-rank-algorithm

Builds a ranking model to predict the relevance score for query-product pairs in HomeDepot’s product search.
https://github.com/pngo1997/learning-to-rank-algorithm

feature-engineering information-retrieval inverted-index learning-to-rank linear-regression mse neural-network point-wise python r-squared ranking-algorithm support-vector-regression text-processing tf-idf xgboost

Last synced: 2 months ago
JSON representation

Builds a ranking model to predict the relevance score for query-product pairs in HomeDepot’s product search.

• Host: GitHub
• URL: https://github.com/pngo1997/learning-to-rank-algorithm
• Owner: pngo1997
• Created: 2024-03-22T19:43:41.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2025-01-31T19:36:48.000Z (8 months ago)
• Last Synced: 2025-02-28T14:13:31.190Z (7 months ago)
• Topics: feature-engineering, information-retrieval, inverted-index, learning-to-rank, linear-regression, mse, neural-network, point-wise, python, r-squared, ranking-algorithm, support-vector-regression, text-processing, tf-idf, xgboost
• Language: Jupyter Notebook
• Homepage:
• Size: 19.7 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# 🛍️ HomeDepot Product Search Relevance Prediction  

## 📜 Overview  

This project builds a **ranking model** to predict the **relevance score** for **query-product pairs** in HomeDepot’s product search. Using **Learning to Rank (LTR)**, we apply a **Pointwise Approach** to train a regression model based on **text similarity features** between the user query and product information.  

📌 **Note**: 

Product description dataset is very large. Please contact me if you want to use it. 

📌 **Dataset**:  

- **Train Set (`train_new.csv`)** – Query-product pairs with **ground-truth relevance scores**.  

- **Test Set (`test_new.csv`)** – Query-product pairs for **prediction**.  

- **Product Descriptions (`product_descriptions_new.csv`)** – Additional product details.  

- **Product Attributes (`attributes_new.csv`)** – Additional structured product attributes.  

📌 **Goal**:  

1. Compute **text similarity** between `search_term` and:  

   - `product_title`  

   - `product_description`  

   - `product_attributes`  

2. Generate **feature vectors** for training and testing.  

3. Train a **machine learning model** to predict **relevance scores**.  

4. Evaluate performance using **Mean Squared Error (MSE) & R² score**.  

📌 **Programming Language**: `Python 3`  

📌 **Libraries Used**: `pandas`, `scikit-learn`, `nltk`, `numpy`, `scipy`, `XGBoost`  

## 🚀 Approach  

### **1️⃣ Data Preprocessing**  

- **Text Cleaning** (e.g., spelling correction, numerical normalization).  

- **Tokenization & Stopword Removal** using `NLTK`.  

- **TF-IDF Vectorization** for product details.  

### **2️⃣ Feature Engineering**  

- Compute **Cosine Similarity** between `search_term` and:  

  - `product_title`  

  - `product_description`  

  - `product_attributes`  

- Compute **additional similarity measures** (e.g., **Jaccard, Dice Coefficient, Overlap**).  

- Minimum **6 similarity features** for each query-product pair.  

### **3️⃣ Model Training & Evaluation**  

- Train models using **Supervised Learning Algorithms**:  

  - **Linear Regression**  

  - **Support Vector Regressor (SVR)**  

  - **XGBoost Regressor**  

  - **Neural Networks**  

- Evaluate model performance using:  

  - **Mean Squared Error (MSE)**  

  - **R² Score**  

### **4️⃣ Predictions on Test Data**  

- Generate **predicted relevance scores** for `test_new.csv`.