Ecommerce Product Categorization

Introduction

This project aims to categorize ecommerce products based on their descriptions. Proper categorization enhances user experience and improves search engine results. The project involves exploratory data analysis (EDA), text normalization, text vectorization, model training, hyperparameter tuning, evaluation, and the development of a Streamlit web application for product category classification.

Product Categories

• Automotive
• Baby Care
• Bags, Wallets & Belts
• Clothing
• Computers
• Footwear
• Home Decor & Festive Needs
• Jewellery
• Kitchen & Dining
• Mobiles & Accessories
• Pens & Stationery
• Tools & Hardware
• Toys & School Supplies
• Watches

Steps Involved

1. Exploratory Data Analysis (EDA):

   • An in-depth analysis of the dataset was conducted to understand its structure and characteristics. Visualizations were created to identify trends, distributions, and patterns within the data.

2. Text Normalization:

   • Text normalization techniques were applied to both the training and test data. Techniques included:
     • Converting text to lowercase
     • Removing whitespace, punctuation,emojis and special characters
     • Lemmatization
     • Removing stopwords
     • Handling HTML tags and hyperlinks
     • POS Tagging
   • The text_normalizer function was used to preprocess the text data.

3. Text Vectorization:

   • Text data was vectorized using TF-IDF vectorization.
   • Both the training and test data were transformed into TF-IDF matrix representations.

4. Model Training:

   • Nine machine learning models were trained to classify products into predefined categories. The RidgeClassifier performed particularly well.

5. Hyperparameter Tuning:

   • Hyperparameters of the best performing model were fine-tuned to optimize performance.

6. Evaluation:

   • The trained model was evaluated on unseen data to assess its performance. An accuracy of 85% was achieved.

7. Streamlit Web Application:

   • A Streamlit web application was developed to allow users to input product descriptions and classify them into relevant categories.

Project Resource Structure

• Datasets: Contains the dataset used for training and testing.
• product-categorization: Jupyter notebooks containing EDA, model training, and evaluation.
• main: Source code for text normalization, vectorization, model training, and the Streamlit web application.
• EDA and Evaluation.ppt: Project report ppt on EDA and Evaluation results.