Promoting the green transformation of the construction industry requires an in-depth understanding of the core elements of various national green building standards. However, the data heterogeneity of standard texts and the scarcity of high-quality annotated data limit the effectiveness of traditional supervised and single unsupervised keyword extraction methods. To address this, this study proposes an unsupervised keyword extraction method with multi-indicator integration, named EBert-MPI. This method integrates semantic indicators, term frequency indicators, and keyword network indicators, and introduces the entropy weight method to achieve objective weighting, enabling a multi-dimensional, comprehensive evaluation of keyword importance. Performance verification shows that EBert-MPI outperforms baseline methods, confirming its comprehensive effectiveness in keyword identification accuracy and importance ranking. Applying EBert-MPI to a cross-national comparative analysis of green building standards from China (ASGB), the United States (LEED), the United Kingdom (BREEAM), and Germany (DGNB) reveals the following results: at the macro level, LEED and ASGB exhibit a relatively balanced focus on the three sustainability dimensions of environment, society, and economy, whereas BREEAM and DGNB concentrate significantly on environmental issues; at the micro level, LEED emphasizes renewable energy and market-driven incentive mechanisms, ASGB uniquely highlights thermal comfort and performance optimization, BREEAM focuses on the standardized management of the construction process, and DGNB advocates for the principle of whole-life-cycle transparency. Meanwhile, the term “design” holds an extremely high and consistent importance ranking across all standards, indicating that integrating sustainability into front-end design has become an internationally recognized core technical paradigm.