Grammar learning has been a bottleneck problem for a long time. In this paper, we propose a method of seman- tic separator learning, a special case of grammar learning. The method is based on the hypothesis that some ...Grammar learning has been a bottleneck problem for a long time. In this paper, we propose a method of seman- tic separator learning, a special case of grammar learning. The method is based on the hypothesis that some classes of words, called semantic separators, split a sentence into sev- eral constituents. The semantic separators are represented by words together with their part-of-speech tags and other infor- mation so that rich semantic information can be involved. In the method, we first identify the semantic separators with the help of noun phrase boundaries, called subseparators. Next, the argument classes of the separators are learned from cor- pus by generalizing argument instances in a hypernym space. Finally, in order to evaluate the learned semantic separators, we use them in unsupervised Chinese text parsing. The exper- iments on a manually labeled test set show that the proposed method outperforms previous methods of unsupervised text parsing.展开更多
目的随着电影内容的复杂化与多样化,电影场景分割成为理解影片结构和支持多媒体应用的重要任务。为提升镜头特征提取和特征关联的有效性,增强镜头序列的上下文感知能力,提出一种混合架构电影场景分割方法(hybrid architecture scene seg...目的随着电影内容的复杂化与多样化,电影场景分割成为理解影片结构和支持多媒体应用的重要任务。为提升镜头特征提取和特征关联的有效性,增强镜头序列的上下文感知能力,提出一种混合架构电影场景分割方法(hybrid architecture scene segmentation network,HASSNet)。方法首先,采用预训练结合微调策略,在大量无场景标签的电影数据上进行无监督预训练,使模型学习有效的镜头特征表示和关联特性,然后在有场景标签的数据上进行微调训练,进一步提升模型性能;其次,模型架构上混合了状态空间模型和自注意力机制模型,分别设计Shot Mamba镜头特征提取模块和Scene Transformer特征关联模块,Shot Mamba通过对镜头图像分块建模提取有效特征表示,Scene Transformer则通过注意力机制对不同镜头特征进行关联建模;最后,采用3种无监督损失函数进行预训练,提升模型在镜头特征提取和关联上的性能,并使用Focal Loss损失函数进行微调,以改善由于类别不平衡导致的精度不足问题。结果实验结果表明,HASSNet在3个数据集上显著提升了场景分割的精度,在典型电影场景分割数据集MovieNet中,与先进的场景分割方法相比,AP(average precision)、mIoU(mean intersection over union)、AUC-ROC(area under the receiver operating characteristic curve)和F1分别提升1.66%、10.54%、0.21%和16.83%,验证了本文提出的HASSNet方法可以有效提升场景边界定位的准确性。结论本文提出的HASSNet方法有效结合了预训练与微调策略,借助混合状态空间模型和自注意力机制模型的特点,增强了镜头的上下文感知能力,使电影场景分割的结果更加准确。展开更多
文摘Grammar learning has been a bottleneck problem for a long time. In this paper, we propose a method of seman- tic separator learning, a special case of grammar learning. The method is based on the hypothesis that some classes of words, called semantic separators, split a sentence into sev- eral constituents. The semantic separators are represented by words together with their part-of-speech tags and other infor- mation so that rich semantic information can be involved. In the method, we first identify the semantic separators with the help of noun phrase boundaries, called subseparators. Next, the argument classes of the separators are learned from cor- pus by generalizing argument instances in a hypernym space. Finally, in order to evaluate the learned semantic separators, we use them in unsupervised Chinese text parsing. The exper- iments on a manually labeled test set show that the proposed method outperforms previous methods of unsupervised text parsing.
文摘目的随着电影内容的复杂化与多样化,电影场景分割成为理解影片结构和支持多媒体应用的重要任务。为提升镜头特征提取和特征关联的有效性,增强镜头序列的上下文感知能力,提出一种混合架构电影场景分割方法(hybrid architecture scene segmentation network,HASSNet)。方法首先,采用预训练结合微调策略,在大量无场景标签的电影数据上进行无监督预训练,使模型学习有效的镜头特征表示和关联特性,然后在有场景标签的数据上进行微调训练,进一步提升模型性能;其次,模型架构上混合了状态空间模型和自注意力机制模型,分别设计Shot Mamba镜头特征提取模块和Scene Transformer特征关联模块,Shot Mamba通过对镜头图像分块建模提取有效特征表示,Scene Transformer则通过注意力机制对不同镜头特征进行关联建模;最后,采用3种无监督损失函数进行预训练,提升模型在镜头特征提取和关联上的性能,并使用Focal Loss损失函数进行微调,以改善由于类别不平衡导致的精度不足问题。结果实验结果表明,HASSNet在3个数据集上显著提升了场景分割的精度,在典型电影场景分割数据集MovieNet中,与先进的场景分割方法相比,AP(average precision)、mIoU(mean intersection over union)、AUC-ROC(area under the receiver operating characteristic curve)和F1分别提升1.66%、10.54%、0.21%和16.83%,验证了本文提出的HASSNet方法可以有效提升场景边界定位的准确性。结论本文提出的HASSNet方法有效结合了预训练与微调策略,借助混合状态空间模型和自注意力机制模型的特点,增强了镜头的上下文感知能力,使电影场景分割的结果更加准确。
