Welcome to Big Data Analytics Assignment 3.

The goal of this assignment is to build a product category classification model without using any labeled data, leveraging fixed BERT embeddings and the Self-Training (Pseudo-Labeling) approach to enable the model to learn from unlabeled text.

**⚠️ Important**: You must not use any labeled data from the previous notebook.
This task assumes a fully unlabeled environment, where only product metadata and category names are available.

You are expected to improve the provided code and surpass the baseline performance (0.15194).