Humanitarian Dialogue System with RDF Ontologies

A task-oriented dialogue system in Congolese Swahili for humanitarian assistance in displacement camps. This system integrates Rasa NLU with RDF ontologies to provide semantic query capabilities.

Project Overview

This system helps displaced people and humanitarian workers access information about:

• Health facilities and services
• Water sources
• Food distribution
• Camp locations and facilities
• Security information

The system uses RDF (Resource Description Framework) to create a structured knowledge base that can be queried using SPARQL.

Key Components

1. RDF Ontology: Defines humanitarian domain concepts using OWL/RDF
2. Triple Store: Stores RDF data and provides SPARQL query endpoint
3. Rasa NLU: Natural language understanding for Congolese Swahili
4. Custom Actions: Connect Rasa intents to SPARQL queries

Setup Instructions

1. Install Dependencies

# Install Python requirements
pip install -r requirements.txt

# Additional packages for RDF support
pip install rdflib SPARQLWrapper

2. Set Up Triple Store

We recommend Apache Jena Fuseki as the SPARQL server:

Important: Apache Jena Fuseki 5.4.0 requires Java 17 or later. Please ensure you have the correct Java version installed before proceeding.

# Check your Java version
java -version

# Download and extract Fuseki
wget https://dlcdn.apache.org/jena/binaries/apache-jena-fuseki-5.4.0.zip
unzip apache-jena-fuseki-5.4.0.zip
cd apache-jena-fuseki-5.4.0

# Start Fuseki server (development mode)
./fuseki-server --mem /humanitarian

If you're using an older Java version and cannot upgrade, you can use an older version of Fuseki that is compatible with your Java installation:

# For Java 11, use Fuseki 4.x
wget https://archive.apache.org/dist/jena/binaries/apache-jena-fuseki-4.8.0.zip
unzip apache-jena-fuseki-4.8.0.zip
cd apache-jena-fuseki-4.8.0
./fuseki-server --mem /humanitarian

Or use the provided installation script:

bash scripts/install_and_setup_fuseki.sh

3. Load RDF Data

Convert the DTM DRC - BA North Kivu Excel data to RDF format:

# Convert Excel to RDF
python scripts/excel_to_rdf.py oim_rdc_dtm_mt_nord-kivu_2024_2_public_hdx.xlsx data/humanitarian_data.ttl

# Load initial ontology
curl -X POST --data-binary @ontology/humanitarian_ontology.ttl \
     --header "Content-Type:text/turtle" \
     http://localhost:3030/humanitarian/data

# Load converted data
curl -X POST --data-binary @data/humanitarian_data.ttl \
     --header "Content-Type:text/turtle" \
     http://localhost:3030/humanitarian/data

4. Configure Rasa Actions

Update the SPARQL endpoint in actions/sparql_actions.py if needed:

SPARQL_ENDPOINT = "http://localhost:3030/humanitarian/query"

5. Train the Rasa Model

Before you can run the bot, you need to train the model:

# Train the Rasa model with the default config
rasa train

# Alternatively, use a specific config file
rasa train --config config_1.yml

6. Start Rasa Action Server

rasa run actions

7. Start Rasa Server

rasa run --enable-api --cors "*" #(optional)

8. Test with Rasa Shell

For interactive testing of the dialogue system:

# Basic shell for testing
rasa shell

# Debug mode with NLU details
rasa shell --debug

# Test with a specific model
rasa shell -m models/20240515-123456.tar.gz

Usage Examples

Example Queries in Congolese Swahili

1. Health Facility Query:

   • "Wapi hii ni vituo vya afya katika Goma?"
   • "Nataka kujua vituo vya afya vinavyotoa huduma ya chanjo"

2. Water Source Query:

   • "Chanzo cha maji katika Bulengo?"
   • "Vyanzo vya maji vilivyopo"

3. Camp Information:

   • "Kambi za wakimbizi wapi?"
   • "Kambi za wakimbizi katika Nyiragongo zina uwezo gani?"

4. Aid Request:

   • "Nahitaji msaada wa chakula"
   • "Ninaomba maji ya kunywa sasa"

SPARQL Query Examples

Example SPARQL query to find health facilities offering vaccination services:

PREFIX : <http://example.org/humanitarian#>
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

SELECT ?facility ?facilityName ?locName
WHERE {
    ?facility rdf:type :HealthFacility .
    ?facility rdfs:label ?facilityName .
    ?facility :hasLocation ?loc .
    ?loc rdfs:label ?locName .
    ?facility :providesService ?service .
    ?service rdfs:label ?serviceName .
    FILTER(REGEX(?serviceName, "Vaccination", "i"))
}

Evaluation

For detailed information about how to evaluate the system, please refer to EVALUATION_README.md. This document describes the evaluation framework, metrics, and provides instructions for:

• Collecting and annotating dialogues
• Running the evaluation scripts
• Visualizing performance results
• Calculating task success metrics

The evaluation framework includes tools to measure task completion rate, dialogue efficiency, and response accuracy.

Project Structure

.
├── actions/                      # Rasa custom actions
│   ├── actions.py                # Standard actions
│   ├── ontology_population.py    # Ontology data loading
│   ├── preprocessing.py          # Text preprocessing
│   ├── query_builder.py          # SPARQL query construction
│   └── sparql_actions.py         # SPARQL query execution
├── data/                         # Training data, RDF data
│   ├── humanitarian_data.ttl     # RDF data
│   ├── nlu_rdf.yml               # NLU training data for RDF
│   ├── rules_rdf.yml             # Rules for RDF responses
│   └── train/                    # Training data split
├── ontology/                     # RDF ontology definitions  
│   └── humanitarian_ontology.ttl # Main ontology file
├── scripts/                      # Utility scripts
│   ├── excel_to_rdf.py           # Excel to RDF converter
│   └── install_and_setup_fuseki.sh # Fuseki setup script
├── tests/                        # Test configuration
├── annotated_dialogues.json      # Evaluation dialogue data
├── config.yml                    # Rasa configuration
├── domain.yml                    # Domain definition
├── EVALUATION_README.md          # Evaluation documentation
├── raw_dialogues.txt             # User-bot conversation samples
├── requirements.txt              # Python dependencies
└── README.md                     # This documentation

Contributing

Feel free to contribute to this project by:

1. Adding more domain concepts to the ontology
2. Improving the Congolese Swahili NLU data
3. Developing additional SPARQL queries
4. Enhancing the dialogue flow

License

This project is open source under the MIT license.