<p>This paper introduces mpA_uvS, an unsupervised video summarization model that leverages multi-perspective attention to evaluate the importance of video frames. The model integrates information from global, local, and spatiotemporal contexts to capture contextual semantic dependencies between video frames. Global attention captures dependencies across the entire video, while local attention focuses on segments within the video. Spatiotemporal attention examines frames at the same spatial positions to identify temporal changes in important objects. By incorporating uniqueness and diversity metrics, mpA_uvS generates summaries that are representative and non-redundant. Experimental results on the SumMe and TVSum datasets demonstrate that mpA_uvS outperforms state-of-the-art unsupervised models and remains competitive with supervised approaches, while achieving efficient training times. The code and datasets are openly available at <a href="https://github.com/cqun-XIN/mpA_uvS">https://github.com/cqun-XIN/mpA_uvS</a></p>