ITEM CLASSIFICATION

Project Overview

This project aims to streamline item classification using advanced NLP and computer vision techniques, making it easier to automate categorization tasks efficiently. The objective is to conduct a feasibility study for an item classification engine based on textual descriptions and/or images.

Data Access

The dataset for this project is available at this address.

Project Objectives

To achieve item classification, the following steps will be undertaken:

1. Dataset Analysis

   • Preprocessing of product descriptions and images.
   • Dimensionality reduction.
   • Clustering of items.
   • Visualization of results in two-dimensional graphs.
   • Similarity calculations (e.g., Adjusted Rand Index - ARI) to compare real categories with generated clusters.

2. Text-Based Classification

   • Implementing different text feature extraction techniques:
     • Bag-of-Words (simple word counting and TF-IDF).
     • Word/Sentence embedding using Word2Vec.
     • Word/Sentence embedding using BERT.
     • Word/Sentence embedding using Universal Sentence Encoder (USE).

3. Image-Based Classification

   • Extracting image features using:
     • Traditional algorithms such as SIFT.
     • CNN-based Transfer Learning approaches.

Project Structure

The project consists of three main notebooks:

• text-preprocessing: Handles text data preprocessing.
• text-clustering-and-classification: Performs item clustering and classification using text data.
• image-clustering-and-classification: Conducts item clustering and classification using images.

Tools & Technologies

• NLP Libraries: NLTK, Word2Vec, BERT, USE
• Machine Learning: Scikit-learn, TensorFlow, Hugging Face
• Computer Vision: OpenCV, VGG-16
• Visualization: Matplotlib, Seaborn