Linking text to SDGs with Artificial Intelligence: Building a multi-use recommender system

UCL Computer Science builds AI-driven SDG labeling tool in collaboration with Knowledge 4 All Foundation

UCL Computer Science has built a cutting-edge AI tool capable of quantifying the link between a piece of text and the SDGs without any expert intervention (report of performance of the tool available upon request). The tool, born from the newly funded research group on AI for Sustainable Development (associated to a new UCL MSc program) has largely surpassed the capabilities of the tool currently used by UCL for this purpose (which was built by the AI SDG Lab), by using the same OSDG dataset but more cutting-edge AI techniques (deep learning models pre-trained on a large Internet corpus). The tool also brings additional functionalities, such as labelling multiple SDGs associated with a text, and providing a probability value (for example: “This piece has a 93% match probability with SDG 4 and a 53% match with SDG 5”). 

The project leverages the power of cutting-edge AI-based language models. These models are usually trained on the whole internet before being fine-tuned on a task (such as SDG tagging). As such, they bring an enormous level of expertise that could reduce the bias in expert labels, save time in expert querying, as well as represent the interconnectedness of the SDGs.  

The final objective for the project is to build an online tool (web app and API) through collaboration with the Knowledge 4 All Foundation. The project team anticipates that with time, the tool can have a wide range of use cases in research and education, not limited to: 

• Automatic mapping of the contribution to the SDGs from a corpus of text resources, being these teaching materials or research papers, which would facilitate creating SDG progress reports around the globe and help to identify gaps in our progress.
• Connecting researchers that work on related SDGs, building stronger communities that work towards common SDGs.
• Recommending open educational materials to students, specifically resources that are related to the SDGs covered in a module. This would strengthen efforts on education for sustainable development, while using the power of hundreds of thousands of open and free educational resources from different learning styles, languages and cultures. 

Work is underway to make the AI model and dataset open-source for the community in the near future. More thorough testing in wider applications and educational settings is to be done in collaboration with the Knowledge 4 All Foundation. There is also intention to focus on applications related to research, enabling researchers that work on similar SDGs to connect. 

The code and dataset for the project will be made available soon. 

---