This project is a graph analysis tool that visualizes relationships between political figures, organizations, and institutions in Chile based on public news articles.

I built this project because I was intrigued by the investigative work behind the Panama Papers (ICIJ). Reading about that investigation, I learned that they didn't just read millions of documents; they treated them as a network. They used graph databases to map how politicians, shell companies, and lawyers were connected.

I wanted to understand how that technology works on a practical level. My goal was to replicate that logic on a smaller scale using Chilean news (Emol) to answer a simple question: If we treat the news as a dataset, who is really connected to whom?

This project takes unstructured text (daily news) and converts it into a structured network, allowing us to see clusters of influence that aren't obvious when reading a single article.

The project runs in three distinct stages:

1. Data Collection: I wrote a script to navigate the site maps of Emol.com and download recent articles. I focused on the text content while filtering out noise like ads and video captions.
2. Entity Extraction: Using Natural Language Processing (spaCy), the system scans every article to identify "Entities" (People and Organizations).
3. Network Construction: If two entities appear in the same article, the system assumes a relationship. The more often they appear together, the stronger the connection (weight) becomes.
4. Visualization: The final output is an interactive web dashboard where nodes are sized by their centrality (importance in the network) and grouped by their connections.

• src/scraper.py: Handles the web scraping logic. It respects server limits and cleans the raw HTML.
• src/processor.py: The core logic. It loads the language model, extracts entities, applies normalization rules, and builds the edge list.
• src/app.py: A Streamlit dashboard that renders the graph using PyVis.
• src/config.py: Contains the configuration settings, including the entity mapping dictionary and stop-words.

• Python 3.10+
• spaCy: For Named Entity Recognition (Spanish large model).
• BeautifulSoup4: For parsing HTML.
• NetworkX & PyVis: For calculating graph metrics and rendering the visualization.
• Streamlit: For the web interface.

1. Clone the repository

git clone https://github.com/kmoragap/emol-graph-project.git
cd emol-graph-project

2. Install dependencies

pip install -r requirements.txt
python -m spacy download es_core_news_lg

3. Run the pipeline This project is modular. You must run the data collection and processing before launching the app.

# 1. Download the news
python src/scraper.py

# 2. Process the text and build the graph
python src/processor.py

# 3. Launch the visualization
streamlit run src/app.py

If you are running this in Google Colab, you will need to use a tunneling service (like localtunnel) to view the Streamlit dashboard, as Colab runs on a remote server. The code in src/app.py is optimized to generate inline resources to support this.