Arango DB Hackathon Submission

Green Living Graph Based Agentic App Project

• Web Apps : https://adbh.pages.dev
• Presentation: Presentation
• Youtube video : https://youtu.be/UNBXhglRGzs

apps.mp4

Table of Contents

• Arango DB Hackathon Submission
• Green Living Graph Based Agentic App Project
  • Table of Contents
  • Project Overview
  • Data Sources
  • Data Processing and Graph Construction
  • Agentic App Functionality

Project Overview

The Green Living Data Analysis Project aims to promote sustainable living by integrating various environmental datasets into a comprehensive knowledge graph. This project leverages data from satellite imagery, mapping services, and news articles to provide insights into green infrastructure, pollution levels, and related news.

Data Sources

• Sentinel Copernicus Satellite Imagery: Provides high-resolution Earth observation data for environmental monitoring.
• OpenStreetMap (OSM): An open-source mapping platform offering detailed information on geographical features, including green spaces and infrastructure.
• Event Registry: Aggregates global news articles, enabling analysis of environmental events and trends.

Data Processing and Graph Construction

1. Geospatial Data Conversion:

   • Converted geospatial data into efficient formats such as Parquet and GeoPackage (GPKG) to optimize storage and access.
   • Performed spatial joins to integrate various geospatial datasets, aligning features based on spatial relationships.

2. News Data Processing:

   • Applied Named Entity Recognition (NER) techniques to extract entities like organizations, locations, and environmental terms from news articles.
   • Utilized Large Language Models (LLMs) to contextualize these entities, linking them to existing graph nodes and uncovering new relationships.

3. Graph Database Schema:

   • Nodes:

     • Object: Represents entities such as Power Generators, EV Charging Stations, Greenery Lands, Public Transport Stations, and Waste Recycle Facilities.
     • ObjectType: Categorizes objects into specific types (e.g., solar power generator, park).
     • Country and City: Geographical entities with associated attributes.
     • Grid: Represents population density and gaseous pollutant levels (CO, CH₄, NO₂).
     • News: Contains news articles with attributes like content and date.
     • NewsEntity and NewsEntityType: Extracted entities from news articles and their classifications.

   • Edges:

     • City → Country: located_in
     • Object → City: located_in
     • Grid → City: located_in
     • Object → ObjectType: is_a
     • News → NewsEntity: mentions
     • News → City: related_to
     • News → Country: related_to
     • NewsEntity → NewsEntityType: belongs_to

Agentic App Functionality

The Agentic App dynamically retrieves and processes natural language queries based on user intent. It offers:

• Geospatial Data Analysis: Provides insights into the distribution and accessibility of green infrastructure and pollutant levels.
• News Retrieval: Aggregates and analyzes news related to environmental issues to keep communities informed.

Example Queries:

• "Find EV charging stations in Berlin."
• "How many greenery lands are in Hamburg?"
• "Show me the location with the highest CO level in Bayern."

Here how it's works