Sustainable Development Goals (SDG) 4.7 aims to ensure learners acquire the knowledge and skills for promoting sustainable development by 2030. Yet, Open Educational Resources (OERs) that connect the public with SDGs are currently limitedly assigned and insufficient to promote SDG and sustainability education to support the achievement of SDG 4.7 and other SDGs by 2030, indicating a need for automatic classification of SDG-related OERs. However, most existing labeling systems can not support multiple labeling, tend to generate a large number of false positives, and have poor transferability within the OER domain. This research proposes a method to automatically assign SDGs based on AutoGluon, a machine-learning framework with powerful predictive capabilities, to allow multiple SDGs to be assigned to each OER. In the proposed framework, challenges of category imbalance and limited data availability are addressed, enhancing the precision and applicability of SDG integration in educational resources. To validate the transferability of model knowledge within the OER corpus, we used 900 lecture video descriptions from SDG Academy, forming the foundation for comparing our framework with existing labeling systems. According to the experiment results, our model demonstrates outstanding merits across various metrics, including precision, recall, F1, ACC, AUC, and AP.

可持续发展目标 (SDG) 4.7 旨在确保学习者到 2030 年获得促进可持续发展的知识和技能。然而，将公众与​​可持续发展目标联系起来的开放教育资源 (OER) 目前分配有限，不足以促进可持续发展目标和可持续发展教育支持到2030年实现可持续发展目标4.7和其他可持续发展目标，表明需要对与可持续发展目标相关的开放教育资源进行自动分类。然而，大多数现有的标记系统不能支持多重标记，往往会产生大量误报，并且在 OER 领域内的可移植性较差。本研究提出了一种基于 AutoGluon（一种具有强大预测能力的机器学习框架）自动分配 SDG 的方法，允许为每个 OER 分配多个 SDG。在拟议的框架中，解决了类别不平衡和数据可用性有限的挑战，提高了可持续发展目标在教育资源中整合的准确性和适用性。为了验证 OER 语料库中模型知识的可转移性，我们使用了 SDG Academy 的 900 个讲座视频描述，为将我们的框架与现有标签系统进行比较奠定了基础。根据实验结果，我们的模型在准确率、召回率、F1、ACC、AUC 和 AP 等各种指标上都表现出了突出的优点。