OntoConnectLM

This code is part of the ANR project SMD. It is supported by IRT SystemX

OntoConnectLM is an open source Python library with a web application demo that aims to assist both industrial domain experts and knowledge/AI engineers in the design, enrichment, and evaluation of ontologies.

Main features

OntoConnectLM provides end-users with features to:

• Dicover concepts from user's textual entries
• Predict relations and RDF triples
• Enrich the ontology concepts with external knowledge graphs annotations (Wikidata and DBpedia)
• Generate an ontology file
  • Populated version
  • Enriched version
• Evaluate the resulted ontology using semantic metrics

Installation and Deployment Guide

Two installation options are available: a local setup and a Docker based setup.

NOTE: You need the Docker based setup to the Web UI.

Local Installation

1. Clone this projet

git clone https://github.com/IRT-SystemX/OntoconnectLM.git 
cd OntoconnectLM

2. Install python dependencies

# create a Python virtual environment
python3 -m venv ./venv
source ./venv/bin/activate

# Install python packages
pip install -r requirements.txt

# Install spacy languages
python -m spacy download fr_core_news_lg && python -m spacy download fr_core_news_sm

3. Run notebooks:

Navigate to the examples/ folder and open the Jupyter notebooks.

Docker Installation

# Go to OntoconnectLM repo.
cd OntoconnectLM

# Build docker image.
docker build --rm -f ./docker/Dockerfile -t smd:latest .

Run notebooks in docker

# On your local terminal
docker run --rm -it -p 8501:8501 -v ${PWD}:/ontoconnectlm smd:latest /bin/bash

# On the running docker container terminal
cd /ontoconnectlm/examples
jupyter notebook --allow-root

Than you need to open the link of jupyternotebook with its token in your favorite web browser.

Web UI

OntoConnectLM provides a Streamlit application that allows users to easily interact with the OntoConnect library.
Please follow the instructions here to run the Web UI.

LLM setup

Ontoconnectlm is available with any llm instance implementing the Langchain BaseLLM interface.

Langchain is available with a lot of LLM providers, including Ollama, Huggingface, MistralAI, OpenAI.

Example with Ollama

from langchain_ollama import OllamaLLM

base_url = "ollama_base_url:11434"
model_name = "gemma3:12b"

llm = OllamaLLM(base_url = base_url, model=model_name) 

Example with Huggingface

from langchain_huggingface import HuggingFaceEndpoint

repo_id = "google/gemma-3n-E4B-it"

llm = HuggingFaceEndpoint(
    repo_id=repo_id,
    max_length=128,
    temperature=0.5,
    huggingfacehub_api_token=HUGGINGFACEHUB_API_TOKEN,
    provider="auto",  # set your provider here hf.co/settings/inference-providers
)

Example with OpenAI

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="gpt-4o",
    temperature=0,
    max_tokens=None,
    timeout=None,
    max_retries=2,
    # api_key="...",  
    # base_url="...",
    # organization="...",
    # other params...
)

LLM instance usage

Once your llm is constructured, you can directly pass it as an argument in Ontoconnectlm objects :

from ontoconnectlm.classes_generator import ClassesGenerator

generator = ClassesGenerator(
    llm = llm,
    ...
)

Examples of usage

1. Test Classes Generator

• Objective: Generate ontology classes based on a given domain context and associated competency questions.

• Steps:

  • Configure LLM object (see section LLM setup)
  • Add your texts describing the domain:
  • Add additional parameters (Optional)
    • Describe the target domain for which classes could be generated.
    • List of competency questions (optional): Enumerate any competency questions that the ontology should address.

• Example:

      Input = ["""Date and time: 20/04/2012 at 07:11, Impact(s) and action(s) taken: Frequency oscillation detected at the France-Belgium border by IDE.""" ,
  
      """Date and time: 08/07/2003 at 08h40
      Centre: Essonne
      Structures concerned: Long outage of the Sevran-Beudottes ACR at the WISSOUS 225 kV substation (16 MW for 5 minutes) following tripping of the 225 kV MASSY-PALAISEAU line.
      Impact(s): Cf. MIN."""],
  
      #Output_list: [Date, Time, 'Event', 'Action', 'Impact', 'Location', 'Client', 'Border']

2. Test Triplet Extractor

• Objective: Extract subject-predicate-object triplets from text using a specified LLM model.
• Input: list of texts, Output: list of triples
• Steps:
  • Configure LLM object (see section LLM setup)
  • Add your texts from which you wish to extract information:
• Example:

  Input = ["""Date and time: 20/04/2012 at 07:11, Impact(s) and action(s) taken: Frequency oscillation detected at the France-Belgium border by IDE.""" ,
  
      """Date and time: 08/07/2003 at 08h40
      Centre: Essonne
      Structures concerned: Long outage of the Sevran-Beudottes ACR at the WISSOUS 225 kV substation (16 MW for 5 minutes) following tripping of the 225 kV MASSY-PALAISEAU line.
      Impact(s): Cf. MIN."""],
  
   #Output_list: Frequency oscillation detection (Event) occuredAtTime 20/04/2012 at 07:11am (DateTime), Frequency oscillation detection (Event) occuredAtPlace Belgium (Location), etc...

3. Test OWL Generator

• Objective: Generate an ontology file populated with previously extracted instances.
• Input: List of triples, Output: OWL/RDF file
• Steps:
  • Indicate a list of triples
  • Specify the classes and properties to be included in the ontology -Example:

  Input = [
          {'label': 'occuredAtTime',
          'head': "Frequency oscillation detection",
          'head_type': 'Event',
          'tail': '20/04/2012 at 07:11am',
          'tail_type': 'Datetime'}
          ]
  
  Output = 
  <owl:ObjectProperty rdf:about="#occuredattime"/>
  ....
  <owl:NamedIndividual rdf:about="#Datetime180">
  <rdf:type rdf:resource="#Datetime"/>
  <rdfs:label rdf:datatype="http://www.w3.org/2001/XMLSchema#string">20/04/2012 at 07h11 am</rdfs:label>
   <has_text rdf:datatype="http://www.w3.org/2001/XMLSchema#string">20/04/2012 at 07h11 am</has_text>
  <   /owl:NamedIndividual>
  ....
  <owl:NamedIndividual rdf:about="#Event127">
  <rdf:type rdf:resource="#Event"/>
  <occuredattime rdf:resource="#Datetime180"/>
  <occuredatplace rdf:resource="#GeographicalRegion10"/>
  <rdfs:label rdf:datatype="http://www.w3.org/2001/XMLSchema#string">Frequency oscillation    detection</rdfs:label>
  <has_text rdf:datatype="http://www.w3.org/2001/XMLSchema#string">Frequency oscillation detection</has_text>
  </owl:NamedIndividual>
   The output file should contain: 'occuredattime' as an owl:objectproperty, 'Datetime180' and 'Event127' as owl:NamedIndividual.

4. Test Onto Enricher

• Objective: Enrich an existing ontology by adding external knowledge from DBpedia and Wikidata.

  • Steps:

    • Load your ontology by getting its content and, and pass it to the enricher service

  • Output example :

  Found text	Wikidata Label	Confidence Score
  Division : Hauts-de-France	Hauts-de-France (French region)	0.9987
  Massive outage occuring on ...	Outage (Electrical event)	0.989
  ... the welding process must ...	Welding (machining)	0.901

5. Test Onto Evaluator

• Objective: Evaluate the quality and consistency of an ontology.

• Steps:

  • Load your ontology by getting its content and, and pass it to the enricher service

• Output example:

  Criterion	Importance Level	Sucess	Explanation
  Presence of disjointness	MEDIUM	FAILED	No disjointness axioms
  Domain and range of properties	MAJOR	PASSED	Every property has a valid domain and range
  Ontology must be available in HTML format	MINOR	PASSED

Local Deployment of DBpedia Spotlight (French)

To enrich your ontologies with French DBpedia (Test Onto Enricher), you can deploy the DBpedia Spotlight service locally using Docker.

Docker Command

docker run -d -p 2222:80 dbpedia/dbpedia-spotlight spotlight-serve fr

How to configure DBpedia Spotlight endpoint in your code

To use your local DBpedia Spotlight service edit the dbpedia_entity_linking as follows:

nlp.add_pipe('dbpedia_spotlight', config={'dbpedia_rest_endpoint': 'http://localhost:2222/rest'})

If you use another port or host, adjust the URL accordingly.

Official Documentation

• Docker Hub: https://hub.docker.com/r/dbpedia/dbpedia-spotlight
• DBpedia Spotlight Documentation: https://www.dbpedia-spotlight.org/installation/

Test Example

To check if the service is running, open in your browser:
http://localhost:2222/rest

Or use curl:

curl "http://localhost:2222/rest/annotate?text=Paris&confidence=0.5"

Licence

This project is under MPL licence.