News-article clustering + deep-learning NLP comparison

Two connected pieces for the Text Analytics module. First, an end-to-end pipeline pulling tweets through the Twitter API, scraping the BBC and CNN articles those tweets linked to, then representing and clustering the resulting text with TF-IDF and K-means to auto-categorize the articles by topic without any manual labeling, evaluated with scikit-learn's clustering metrics and Yellowbrick's visual diagnostics, with word clouds and matplotlib plots to inspect the clusters directly rather than trusting a single score.

Second, a separate comparative study sitting alongside the clustering work: BERT, XLNet, and GPT-3 evaluated against the same underlying classification problem, comparing how three different eras of language-model architecture handled it.

Doing the practical clustering pipeline and the comparative deep-learning study inside one module meant the "classical NLP versus modern transformer models" question wasn't abstract, the classical TF-IDF/K-means approach and the transformer comparison ran against text pulled from the same real source.