Unlocking Global Impact: Automating Research Classification for Sustainable Development Goals

      The world faces urgent challenges, from climate change and poverty to healthcare access and sustainable energy. The United Nations’ 2030 Agenda for Sustainable Development, with its seventeen Sustainable Development Goals (SDGs), provides a universal roadmap to address these issues. Achieving these ambitious goals hinges significantly on scientific research and innovation, which provide the foundational knowledge for policy, technological advancements, and societal transformation.

      However, as the volume of scholarly publications explodes across diverse disciplines, systematically evaluating how this vast body of research contributes to the SDGs has become an immense challenge. Manual classification, performed by human experts, is not only impractical due to the sheer number of articles but also time-consuming, costly, and prone to inconsistencies or subjective bias. This bottleneck highlights a critical need for automated, scalable, and transparent solutions to consistently map research outputs to relevant SDG categories.

The Challenge of Mapping Research to Global Goals

      The sheer scale of academic output today makes human-driven analysis an insurmountable task. Universities, funding organizations, and policymakers are increasingly keen to understand and demonstrate their contributions to global sustainability. Yet, without an efficient way to categorize research, much of this impact remains obscured. Traditional methods struggle to keep pace with the influx of new knowledge, leading to difficulties in identifying trends, allocating resources effectively, or tracking progress towards sustainability targets. This lack of systematic oversight means that valuable insights embedded in research might not be fully leveraged to inform critical decisions.

      Existing approaches to linking research with SDGs often fall into two main categories: rule-based or machine learning-based. Rule-based methods rely on pre-defined keyword queries, often curated by experts, to identify SDG-relevant content. These queries typically incorporate thematic keywords and action-oriented terminology drawn from official SDG targets and indicators. For example, identifying research related to SDG 1 (No Poverty) might involve searching for combinations of terms like "poverty," "inequality," and "social protection." This approach offers high transparency, allowing every SDG assignment to be explicitly traced back to the keywords found in the research text. Organizations like the Aurora Universities Network and Elsevier have developed extensive SDG mapping query sets, which have become valuable resources for bibliometric analyses and university ranking exercises. The Elsevier 2023 SDG mapping dataset, in particular, is a widely referenced public resource used to reproduce SDG-related research analytics.

Introducing a Transparent AI-Powered Framework for SDG Classification

      To overcome the limitations of manual processes and the opacity of some automated methods, a new computational framework proposes an automated, rule-based model for classifying research papers. This system leverages expert-curated Boolean query mappings, which are structured expressions used to process bibliographic metadata such as titles, abstracts, and keywords. By using a modular, web-based design, it provides an accessible and intuitive platform for a wide range of users.

      The framework is built on a three-tiered architecture: a presentation layer, an integration layer, and a classification engine. The presentation layer offers a user-friendly web interface, built with modern UI libraries, where users can input data for either single-paper classification (manually entering a title, abstract, and keywords) or batch classification (uploading structured files like CSV or TSV with multiple records). Users can customize parameters, such as the number of top SDGs to be returned, and instantly visualize the results in ranked lists and interactive charts. The integration layer, comprising a backend API, handles high-throughput processing, while a Python-based classification engine executes complex Boolean queries. This design allows for rapid and efficient processing of thousands of research records per hour, delivering reproducible and consistent results.

Beyond Keywords: Advantages of a Rule-Driven Approach

      While machine learning (ML) and natural language processing (NLP) models have been explored for SDG classification, offering the ability to capture semantic nuances beyond exact keyword matches, they often come with significant drawbacks. ML models typically require substantial computational resources for training and parameter optimization. More importantly, many machine learning-based systems are "black boxes," meaning their classification decisions are not easily interpretable on a case-by-case basis. This lack of transparency can be a major hurdle in audit-sensitive applications where SDG assignments need clear justification.

      In contrast, the proposed Boolean query-based approach offers unparalleled interpretability. Each SDG assignment is explicitly linked to the specific keywords and query parts that triggered the classification. This transparency makes the system highly auditable and trustworthy, crucial for institutions and policymakers who need to justify their research funding, policy decisions, or public reporting on sustainability efforts. This interpretability ensures that stakeholders can understand why a particular paper was assigned to an SDG, fostering confidence and enabling easier refinement of the classification logic.

Real-World Impact: Classifying Research at Scale

      The ability to process vast datasets quickly and consistently offers substantial benefits. Institutions can gain a systematic understanding of their research portfolios' alignment with global sustainability goals, aiding strategic planning and resource allocation. Researchers can identify relevant literature more efficiently, and policymakers can gather evidence-based insights to formulate informed strategies. The framework’s capacity for both single-paper and batch classification makes it versatile for individual researchers and large organizations alike. It represents a viable solution for anyone interested in systematically analyzing research alignment with sustainability goals, without the complexity and opacity often associated with advanced machine learning models.

      Companies like ARSA Technology leverage similar principles of converting raw data into actionable intelligence across various industries. For instance, in retail environments, ARSA's solutions for Smart Retail Counter use AI to analyze customer flow and behavior, generating insights that optimize store layouts and staffing. Similarly, for urban planning and security, ARSA’s AI Box - Traffic Monitor system processes real-time video data to classify vehicles and detect congestion, providing crucial information for traffic management. These applications demonstrate how intelligent systems, whether rule-based or machine learning-driven, can transform data into tangible business outcomes, much like this framework transforms research metadata into sustainability insights. ARSA's focus on AI Video Analytics across various industries highlights the broad applicability of such AI-powered data processing.

      By providing clear and understandable outputs that reveal the precise query components motivating each SDG assignment, this framework champions explainable AI. Its experimental testing has demonstrated the ability to process thousands of research records in an hour with reproducible and consistent results, proving its efficacy for large-scale application.

      The framework proposes a scalable, transparent, and user-friendly method for automating the classification of research papers against the UN Sustainable Development Goals. It underscores the critical role of interpretable AI in addressing complex global challenges, offering a robust tool for academic institutions, funding bodies, and policymakers to systematically track and promote sustainability-focused research.

      **Source:** Dewani, S., & Sharma, K. (2026). Automated Classification of Research Papers Toward Sustainable Development Goals: A Boolean Query-Based Computational Framework. arXiv preprint arXiv:2601.16988. https://arxiv.org/abs/2601.16988

      To explore how ARSA Technology's AI and IoT solutions can bring actionable intelligence to your operations, we invite you to connect with our experts for a free consultation.