🛍️ 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.