Hackathon Savoirs

Spatial analysis and reading recommendation system for the Savoirs digital humanities corpus.

Overview

Analyzes the Savoirs French academic corpus by extracting geographic references from TEI/XML-encoded articles, geocoding place names, and computing spatial similarity between articles. Articles that discuss geographically related topics are surfaced as reading recommendations.

Tech Stack

• Geospatial: GeoPandas, Shapely, Geopy (Nominatim), Geocoder (GeoNames API)
• XML Processing: BeautifulSoup with lxml parser
• Data Processing: Pandas
• Data Format: GeoPackage (GPKG)

Pipeline

TEI/XML Corpus (~100 articles)
    ↓
Extract place names from <placename> tags
    ↓
Geocode locations (GeoNames API + Nominatim fallback)
    ↓
Save as GeoPackage (one per article)
    ↓
Compute pairwise spatial distances between articles
    ↓
Export similarity rankings as CSV

Project Structure

hackaton_savoirs/
├── python/
│   ├── pre_processing.py        # Geocoding pipeline (TEI → GeoPackage)
│   ├── main.py                  # Pairwise distance computation
│   ├── functions_xml.py         # TEI/XML extraction (titles, authors, places)
│   ├── functions_geodata.py     # Geocoding and GeoDataFrame creation
│   ├── geo_article.py           # GeoArticle class for spatial similarity
│   ├── functions.py             # General utilities
│   └── tests/                   # Unit tests
├── CorpusTEI/                   # Input: TEI-encoded articles (ANABASES journal)
└── data/
    ├── article_locations/       # Output: GeoPackage files per article
    └── distances/               # Output: CSV distance matrices

Usage

# Stage 1: Geocode place names from TEI corpus
python python/pre_processing.py

# Stage 2: Compute pairwise article distances
python python/main.py

Context

Built during the Hackathon Savoirs for challenges on the Savoirs corpus: extracting knowledge through text mining, spatial analysis, and data visualization, and designing reading recommendation strategies based on geographic and thematic similarity.