Technological University Dublin
IL-3: Sustainability Course Inventory

A sustainability language model to classify text among the 17 UN SDGs and identify the sustainability level as Sustainability-Focused, Sustainability-Inclusive, or Non-Sustainable has been developed and is currently being patented. The identification of sustainability challenges and alignment with specific SDGs was based on a robust framework that integrates natural language processing (NLP) and machine learning (ML) capabilities.

 

The framework comprises two main components:

 

Sustainability-Tuned NLP Models: These models leverage NLP and ML to:

Classify text among the 17 UN SDGs,

Assess sustainability importance (or “sustainability score”) with a confidence level, and

Classify sustainability scores according to definitions aligned with the AASHE STARS framework, providing standardized terms like "Sustainability-Focused," "Sustainability-Inclusive," and "Non-Sustainable."

These tailored NLP models, referred to as “sustainability-tuned language models,” or simply “sustainability models,” are specifically designed to categorize sustainability-related content accurately.

 

Supervisory Model for Enhanced Accuracy: An overarching model supervises the outputs of each sustainability model, automatically identifying the most accurate results to consolidate and deliver reliable classifications. This supervisory model thus enhances consistency and precision across the various sustainability models.

Any sustainability-tuned language model or large language model (LLM) can be developed and integrated into this framework. These models and the framework can be applied to analyze text-based media of any type, such as documents, files, online content, and even other models. Applications of this technology are broad and include sustainability assessments for curriculum content, research outputs, websites, and any textual document. Furthermore, the models and framework can be integrated into various platforms, such as software applications, web applications, and additional models.

 

The output of the sustainability models includes:

 

Relevant SDG(s) identified within the text,

Sustainability scores with associated confidence levels for each SDG, and

An overall sustainability score and confidence level that classifies the text as “Not Sustainable,” “Medium Sustainability,” or “High Sustainability.” The latter two categories align with AASHE STARS classifications of “Sustainability Inclusive” and “Sustainability Focused,” respectively.

This model, which boasts an accuracy rate exceeding 86% across all SDGs, has undergone extensive manual review by sustainability experts, including the TU Dublin Sustainability Office and sustainability champions across various departments. In cases with multiple course offerings or sections, the model aggregates data to ensure comprehensive, accurate sustainability assessments.