提出了一种基于情感分布的emoji嵌入式表示方法(emoji embedded representation based on emotion distribution,EDEER)。EDEER方法采用基于BERT的情绪预测模型软标签,从真实数据中学习emoji嵌入式表示,通过情感分布直接建模emoji在各...提出了一种基于情感分布的emoji嵌入式表示方法(emoji embedded representation based on emotion distribution,EDEER)。EDEER方法采用基于BERT的情绪预测模型软标签,从真实数据中学习emoji嵌入式表示,通过情感分布直接建模emoji在各种情绪上的表达程度,使嵌入式表示中包含emoji的多种情感信息。在包含emoji的中文微博数据集上的多组对比实验表明,本文提出的方法可以有效地学习到与细粒度情绪直接关联的emoji嵌入式表示,构建具有较高情绪表达质量的emoji表示空间。展开更多
随着深度学习的发展,越来越多基于神经网络的算法用于实现文本情感分类,在分类上的精度不断提升,如果一味追求精度而加深网络的层次,会给实际应用场景中的响应等性能带来较大阻碍。通过研究文本的嵌入式表示等技术,在时下前沿的FastTex...随着深度学习的发展,越来越多基于神经网络的算法用于实现文本情感分类,在分类上的精度不断提升,如果一味追求精度而加深网络的层次,会给实际应用场景中的响应等性能带来较大阻碍。通过研究文本的嵌入式表示等技术,在时下前沿的FastText模型基础上进一步捕捉分类逻辑中重点的文本特征,提出了新的轻量化的权重驱动的文本情感分类模型WDFT (Weight-Driven Fast Text)。在实现高精度的同时保证模型的轻量化,更好地解决文本情感分类任务。展开更多
Map data display is the basic information representation mode under embedded real-time navigation. After a navigation display data set (NDIS_SET) with several dimensions and corresponding mathematical description fo...Map data display is the basic information representation mode under embedded real-time navigation. After a navigation display data set (NDIS_SET) with several dimensions and corresponding mathematical description formula are designed, a series of rules and algorithms are advanced to optimize embedded navigation data and promote data index and input efficiency. A new parallel display algorithm with navigation data named N PDIS is then presented to adapt to limited embedded resources of computation and memory after a normal navigation data display algorithm named NDIS and related problems are analyzed, N_PDIS can synchronously create two preparative bitmapa by two parallel threads and switch one of them to screen automatically. Compared with NDIS, the results show that N_PDIS is more effective in improving display efficiency.展开更多
文摘提出了一种基于情感分布的emoji嵌入式表示方法(emoji embedded representation based on emotion distribution,EDEER)。EDEER方法采用基于BERT的情绪预测模型软标签,从真实数据中学习emoji嵌入式表示,通过情感分布直接建模emoji在各种情绪上的表达程度,使嵌入式表示中包含emoji的多种情感信息。在包含emoji的中文微博数据集上的多组对比实验表明,本文提出的方法可以有效地学习到与细粒度情绪直接关联的emoji嵌入式表示,构建具有较高情绪表达质量的emoji表示空间。
文摘随着深度学习的发展,越来越多基于神经网络的算法用于实现文本情感分类,在分类上的精度不断提升,如果一味追求精度而加深网络的层次,会给实际应用场景中的响应等性能带来较大阻碍。通过研究文本的嵌入式表示等技术,在时下前沿的FastText模型基础上进一步捕捉分类逻辑中重点的文本特征,提出了新的轻量化的权重驱动的文本情感分类模型WDFT (Weight-Driven Fast Text)。在实现高精度的同时保证模型的轻量化,更好地解决文本情感分类任务。
文摘Map data display is the basic information representation mode under embedded real-time navigation. After a navigation display data set (NDIS_SET) with several dimensions and corresponding mathematical description formula are designed, a series of rules and algorithms are advanced to optimize embedded navigation data and promote data index and input efficiency. A new parallel display algorithm with navigation data named N PDIS is then presented to adapt to limited embedded resources of computation and memory after a normal navigation data display algorithm named NDIS and related problems are analyzed, N_PDIS can synchronously create two preparative bitmapa by two parallel threads and switch one of them to screen automatically. Compared with NDIS, the results show that N_PDIS is more effective in improving display efficiency.