针对基础深度学习模型特征提取能力不足,循环网络训练效率低等问题,将高校社交网络平台评论文本数据作为研究对象,提出了基于多尺度语义协同网络的高校网络舆论情感分类模型。预训练模型ALBERT(A Lite BERT)通过结合当前词的具体上下文...针对基础深度学习模型特征提取能力不足,循环网络训练效率低等问题,将高校社交网络平台评论文本数据作为研究对象,提出了基于多尺度语义协同网络的高校网络舆论情感分类模型。预训练模型ALBERT(A Lite BERT)通过结合当前词的具体上下文动态调整向量表示,提升词向量语义表征质量。多尺度语义协同网络捕捉评论文本不同尺度下的多通道融合情感特征,软注意力机制计算每个特征对情感分类结果的影响权重大小,加权求和后得到情感分类特征表示,线性层输出分布概率并得到具体情感倾向。在真实高校图书馆社交网络平台用户评论数据集进行实验,结果表明该模型F1分数达到了97.46%,优于近期表现优秀的实验对比模型,且通过消融实验证明了各个功能模块的有效性。展开更多
The purpose of this paper is to explore the trade-offs and synergies of multifunctional cultivated land(MCL) at multiple scales. The study area is Wuhan Metropolitan Area, China. The entropy method and the method of S...The purpose of this paper is to explore the trade-offs and synergies of multifunctional cultivated land(MCL) at multiple scales. The study area is Wuhan Metropolitan Area, China. The entropy method and the method of Spearman’s rank correlation were employed for the analysis of combined land use/cover data, administrative division data, population data and statistical yearbook data, from the multi-scale perspectives of cities, counties and townships. The results showed that:(1) The multi-functionality of cultivated land had obvious spatial differences and its overall spatial patterns were relatively robust, which did not change very much at the single scale.(2) At each single scale, the MCL’s trade-offs and synergies had spatial heterogeneity.(3) Scale effects existed in the MCL’s trade-offs and synergies. From the prefecture-level city scale, to the county scale, and to the township scale, the MCL’s trade-offs were changed to synergies, and some synergic relationships were enhanced. This article contributes to the literature by deepening the multiscale analysis of trade-offs and synergies of multifunctional cultivated land. The conclusions might provide a basis for helping policy-makers to implement protection measures for the multi-functionality of cultivated land at the right spatial scale, and to promote the higher-level synergies of multifunctional cultivated land to realize its sustainable use.展开更多
文摘针对基础深度学习模型特征提取能力不足,循环网络训练效率低等问题,将高校社交网络平台评论文本数据作为研究对象,提出了基于多尺度语义协同网络的高校网络舆论情感分类模型。预训练模型ALBERT(A Lite BERT)通过结合当前词的具体上下文动态调整向量表示,提升词向量语义表征质量。多尺度语义协同网络捕捉评论文本不同尺度下的多通道融合情感特征,软注意力机制计算每个特征对情感分类结果的影响权重大小,加权求和后得到情感分类特征表示,线性层输出分布概率并得到具体情感倾向。在真实高校图书馆社交网络平台用户评论数据集进行实验,结果表明该模型F1分数达到了97.46%,优于近期表现优秀的实验对比模型,且通过消融实验证明了各个功能模块的有效性。
基金The National Natural Science Foundation of China (71673105)The Fundamental Research Funds for the Central Universities (2662016PY116)。
文摘The purpose of this paper is to explore the trade-offs and synergies of multifunctional cultivated land(MCL) at multiple scales. The study area is Wuhan Metropolitan Area, China. The entropy method and the method of Spearman’s rank correlation were employed for the analysis of combined land use/cover data, administrative division data, population data and statistical yearbook data, from the multi-scale perspectives of cities, counties and townships. The results showed that:(1) The multi-functionality of cultivated land had obvious spatial differences and its overall spatial patterns were relatively robust, which did not change very much at the single scale.(2) At each single scale, the MCL’s trade-offs and synergies had spatial heterogeneity.(3) Scale effects existed in the MCL’s trade-offs and synergies. From the prefecture-level city scale, to the county scale, and to the township scale, the MCL’s trade-offs were changed to synergies, and some synergic relationships were enhanced. This article contributes to the literature by deepening the multiscale analysis of trade-offs and synergies of multifunctional cultivated land. The conclusions might provide a basis for helping policy-makers to implement protection measures for the multi-functionality of cultivated land at the right spatial scale, and to promote the higher-level synergies of multifunctional cultivated land to realize its sustainable use.
