SEMIC's AI-Powered Chatbot for data modelling

Introduction

In the realm of semantic data modeling, it is essential to define concepts such as classes, properties, relationships, and code lists that constitute a data model. Data models are omnipresent; for instance, public administrations utilize them to interact with citizens by delineating their catalogs of services, including names, descriptions, contact points, and applicable rules.

Professionals engaged in data modeling often seek existing solutions to avoid unnecessary duplication of effort. However, the challenge lies in the difficulty of pinpointing the specific concept that aligns with their needs, often making the search feel like finding a needle in a haystack. This can lead to a substantial investment of time and may dishearten semantic modelers from applying best practices. Additionally, when faced with new elements, data modelers often struggle to generate accurate definitions or lists of values using the correct terminology. Another pressing issue arises in determining whether two elements are truly identical in meaning or merely resemble each other while possessing subtle differences.

To tackle these challenges, SEMIC has developed a chatbot designed to assist semantic modelers by answering their queries and expediting their work. For example, during a project, a modeler may need to incorporate new elements such as classes or relationships into an existing model but might be unsure of where to begin. The SEMIC chatbot offers substantial support in this area.

Through the chatbot interface, users can easily upload their model and begin an interactive dialogue. The chatbot is capable of answering inquiries regarding existing classes, relationships, or properties within the model, which aids in understanding the current structure. Most importantly, it serves as a brainstorming partner, suggesting possible new additions. For instance, if a modeler is developing a vocabulary to describe a person's attributes and related concepts, the chatbot can help identify the need to include a class for a person's medical history. Furthermore, it scans a library of existing models to suggest relevant concepts.

Once a modeler finalizes a new element, the chatbot is equipped to convert it into JSON-LD or to directly integrate it into the existing model. The tool also allows for the visualization of the revised model via the interface. After the modeler has made satisfactory modifications, they can export it in the same format as the original model, which enables smooth integration into their preferred editing software.

The chatbot is powered by the same underlying model as the renowned ChatGPT. However, to enhance its utility for data modeling, the SEMIC team has supplemented it with additional specialized knowledge. This is accomplished using a technique known as Retrieval-augmented Generation (RAG). When a user poses a question, the chatbot searches its knowledge base for pertinent information, integrating general training data with specialized sources. These include seven distinct references, such as established models like CPSV-AP, legislation like the Open Data Directive, and technical documentation, including RDF files from linked open vocabularies and Schema.org.

The new tool is currently in its infancy stage, with the SEMIC team aiming to broaden its knowledge base and introduce features, including the capability to generate a complete data model based on specific requirements. This function would enable the chatbot to create models with definitions, classes, and properties, which could subsequently be visualized in diagrams and exported in both human-readable and machine-readable formats. This positions the tool as a significant asset for data modeling tasks.

As SEMIC continues to explore how AI can facilitate interoperability and improve the working lives of data professionals, policy officers, and public administrations across Europe, they invite interested parties to join in the development of the chatbot and other AI initiatives aimed at enhancing interoperability.

Keywords

• Semantic Data Modeling
• Data Models
• Public Administrations
• AI Chatbot
• Semantic Modelers
• JSON-LD
• Retrieval-augmented Generation (RAG)
• Interoperability

FAQ

1. What is semantic data modeling?
Semantic data modeling is the process of defining the concepts that make up a data model, including classes, properties, relationships, and code lists.

2. How does SEMIC's chatbot assist data modelers?
The chatbot helps data modelers by providing answers to their questions about existing model elements, facilitating brainstorming for new additions, and integrating modifications into the existing model.

3. What technique does the chatbot use to find relevant information?
The chatbot employs Retrieval-augmented Generation (RAG), which allows it to search a knowledge base and provide the most suitable information as an answer to user queries.

4. Can the chatbot visualize updated models?
Yes, the chatbot has features that allow users to visualize the updated model through its interface.

5. What future capabilities are planned for the chatbot?
Future developments include the generation of complete data models based on specific requirements, incorporating definitions, classes, properties, and visualization features.