Data Visualization and Text Mining

Università Cattolica del Sacro Cuore Academic Year 2025-2026

About

This repository contains notebooks and materials for the Data Visualization and Text Mining course. Notebooks will be published the day before each lesson.

Course Contents

01 - NLP In Practice

Introduction to Natural Language Processing using spaCy and NLTK.

• NLP Basics and pipeline overview
• Tokenization, Stemming, Lemmatization
• Stop Words
• Part of Speech (POS) Tagging and visualization
• Named Entity Recognition (NER) and visualization with displacy
• Exploratory Data Analysis on Text Data

02 - Text Classification

Machine Learning applied to text data.

• Scikit-Learn basics
• N-grams
• Bag-of-Words, CountVectorizer, TfidfVectorizer
• Text Classification project (end-to-end)

03 - Neural Networks

Introduction to Deep Learning for NLP.

• Simple Classifier with Keras
• PyTorch Simple Classifier

04 - Embeddings

Word vectors and text representation.

• Feature Engineering for Text Data
• Word2Vec (CBOW and Skip-gram)
• GloVe embeddings
• Sentence-Transformers (modern embeddings)

05 - LSTM

Recurrent Neural Networks for NLP.

• Text Generation with Neural Networks
• LSTM for Sentiment Classification
• LSTM for Named Entity Recognition (BiLSTM on CoNLL-2003)

06 - Topic Modeling

Unsupervised methods for discovering topics in text.

• Latent Dirichlet Allocation (LDA)
• Non-Negative Matrix Factorization (NMF)
• Topic Model Evaluation (coherence, perplexity, diversity)
• BERTopic (modern embedding-based approach)

07 - Data Visualization

Comprehensive guide to data visualization in Python.

• Matplotlib basics and line plots
• Area plots, histograms, and bar charts
• Pie charts, box plots, scatter plots, and bubble plots
• Waffle charts, word clouds, and regression plots
• Generating maps in Python
• Plotly basics for interactive visualizations

08 - Dashboards

Interactive dashboard development with Dash.

• Layout creation with HTML and Dash Bootstrap Components
• Navigation bars and cards
• HTML and core components
• Tables and interactive elements
• Callbacks (basic, multiple inputs/outputs, chained, with State)
• Real-world applications: COVID dashboard, sales app, NLP Q&A app

09 - Transfer Learning

Practical demonstrations of transfer learning in NLP.

• Transfer Learning Demo: Comparing three approaches on BERT
  • Training from scratch (random initialization)
  • Feature extraction (frozen BERT + classifier)
  • Fine-tuning (end-to-end adaptation)
  • Performance comparison and visualization
• ULMFit Experiment: Three-step transfer learning process
  • General-domain language model pre-training (simulated)
  • Target task language model fine-tuning
  • Classifier training with gradual unfreezing
  • Discriminative fine-tuning with layer-wise learning rates

10 - Attention Mechanisms

Understanding attention in neural networks.

• Bahdanau Attention (simplified implementation)
• Visualization of attention weights
• Seq2seq models with attention

11 - Transformers

Introduction to Transformer architecture and Hugging Face ecosystem.

• Transformer anatomy and self-attention mechanism
• Multi-head attention visualization
• Hugging Face Transformers library introduction
• Working with pre-trained models
• Tokenizers and model pipelines

12 - BERT (Encoder Models)

Advanced applications with BERT-based models.

• Text Classification with Transformers
  • Fine-tuning BERT for sentiment analysis
  • Model evaluation and comparison
  • Comprehensive metrics (Accuracy, Precision, Recall, F1, ROC)
• Question Answering with fine-tuned BERT
  • SQuAD dataset and QA task
  • Extractive QA implementation
  • Exact Match (EM) and F1 metrics
  • Confidence scores and answer extraction

13 - GPT (Decoder Models)

Text generation with GPT models.

• Autoregressive text generation
• Sampling strategies (temperature, top-k, top-p)
• Generation metrics
  • Perplexity (model confidence)
  • Diversity metrics (Distinct-n, Self-BLEU)
  • BLEU and ROUGE scores
• BERT vs GPT comparison
• Controlled generation techniques

14 - RAG & Modern LLMs

Real-world applications combining retrieval and generation.

• RAG Pipeline (NLP14_1_RAG_Pipeline.ipynb)
  • Building knowledge bases with vector databases (FAISS)
  • Semantic search with embeddings
  • Retrieval-Augmented Generation architecture
  • Retrieval metrics (Precision@K, Recall@K, MRR, NDCG)
  • Generation quality metrics (faithfulness, relevance)
  • End-to-end performance analysis
• Modern LLM APIs (NLP14_2_Modern_LLMs.ipynb)
  • Working with GPT-4, Claude, and other modern LLMs
  • Prompt engineering techniques
  • Few-shot learning
  • Cost-performance optimization
  • LLM evaluation frameworks
  • Production best practices

15 - AI Agents

Autonomous systems that use LLMs with tools to complete complex tasks.

• ReAct Agents - Reasoning + Acting in iterative loops
• Agent Components
  • LLM brain for decision-making
  • Tool integration (APIs, calculators, databases)
  • Memory systems (short-term and long-term)
  • Multi-step planning and execution
• Practical Implementation
  • Building agents with OpenAI function calling
  • Building agents with Gemini function calling (FREE)
  • Calculator, Wikipedia search, weather, and datetime tools
  • Agent with conversation memory
  • Error handling and best practices

---

🆓 FREE API Access for Students

Many modern LLM notebooks require paid API keys (OpenAI, Anthropic), which can be a barrier for students. We've created free alternatives using Google Gemini that require no credit card!

Gemini-Based Notebooks

Three complete notebooks using 100% FREE Google Gemini API:

1. Modern LLMs with Gemini (14-RAG/NLP14_2_Modern_LLMs_Gemini.ipynb)

   • Prompt engineering, few-shot learning, evaluation
   • No credit card required

2. AI Agents with Gemini (15-AI_Agent/NLP15_1_AI_Agent_Gemini.ipynb)

   • Autonomous agents with tool use
   • Function calling, memory, multi-step reasoning

Getting Your Free Gemini API Key

1. Go to Google AI Studio
2. Sign in with your Google account
3. Click "Get API Key" → "Create API key"
4. Copy your key and use it in the notebooks

Free Tier Limits:

• 60 requests per minute
• 1,500 requests per day
• 1 million tokens per month
• More than enough for all coursework!

---

Getting Started

Clone the Repository

git clone <repository-url>
cd text-mining-dataviz-aa2526

Stay Updated

Since new content is added before each lesson, make sure to pull the latest changes regularly:

git pull origin main

Tip: Run this command before starting a lesson to get the new notebooks.

Open Notebooks in Google Colab

You can open any notebook directly in Google Colab without cloning the repository:

1. Navigate to the notebook file on GitHub
2. Replace github.com in the URL with colab.research.google.com/github

Example:

# Original GitHub URL
https://github.com/nluninja/text-mining-dataviz-aa2526/blob/main/notebook.ipynb

# Colab URL
https://colab.research.google.com/github/nluninja/text-mining-dataviz-aa2526/blob/main/notebook.ipynb

Alternatively, from Google Colab:

1. Go to colab.research.google.com
2. Click File → Open notebook
3. Select the GitHub tab
4. Enter the repository URL and select the notebook you want to open

Note: Changes made in Colab are saved to your Google Drive, not to the repository. To keep your work, save a copy to your Drive.

Workflow for Each Lesson

1. Before the lesson: Pull the latest updates

   git pull origin main

2. During the lesson: Work through the notebooks

3. After the lesson: Your local changes won't conflict with future updates as long as you don't modify the original files. If you want to experiment, consider creating a copy of the notebook.

Handling Conflicts

If you've made changes to a notebook and encounter conflicts when pulling:

# Option 1: Stash your changes, pull, then reapply
git stash
git pull origin main
git stash pop

# Option 2: Discard local changes and get the latest version
git checkout -- .
git pull origin main

Project Track

The project-track/ folder contains information and resources for the final course project:

• README.md - Complete project requirements and guidelines
• datasets/ - Collection of 10 curated datasets for text classification and entity extraction tasks
  • 5 Text Classification datasets (Sentiment Analysis, News Categorization, Spam Detection, etc.)
  • 5 Entity Extraction datasets (NER, Product Attributes, Medical Entities, etc.)
  • LLM_USE_CASES.md - Optional guidance for using LLMs in your project

Project Requirements Summary

Students must build a text processing pipeline that includes:

1. Data Exploratory Analysis with visualizations
2. Neural Network approach (LSTM, BiLSTM, RNN, etc.)
3. Transformer-based approach (BERT, etc.)
4. Model comparison and metrics
5. Interactive dashboard combining all aspects

See the project-track README for full details and submission guidelines.

Tools

The tools/ folder contains helpful resources and tutorials:

• Git-Quick-Tutorial.md - A quick reference guide for Git commands and workflows
• Git-Exercises.md - Practice exercises to reinforce Git skills

License

This project is licensed under the MIT License - see the LICENSE file for details.

This repository is intended for academic and educational purposes.

Contact

For questions about the course materials, please contact the instructor:

Andrea Belli - andrea.belli@unicatt.it