ICLR2015文章《ENSEMBLEOFGENERATIVEANDDISCRIMINATIVETECHNIQUESFORSENTIMENTANALYSISOFMOVIEREVIEWS》指出，在实验中使用层级softmax而非负采样匹配(Le&Mikolov,2014)的结果时，仅当训练和测试数据未被打乱的情况下才能得到92.6%的准确性，因此认为该结果无效。
Quoc V. Le ICML2014的工作《Distributed Representations of Sentences and Documents》一直饱受争议，很多学者都宣称其工作无法重现。
ICLR2015文章《ENSEMBLE OF GENERATIVE ANDDISCRIMINATIVETECHNIQUES FORSENTIMENTANALYSIS OFMOVIEREVIEWS》在文章中直接对该工作进行了评价：
In our experiments, to match the results from (Le & Mikolov, 2014), we followed the suggestion by QuocLe to use hierarchical softmax instead of negative sampling. However, this produces
the 92.6% accuracy resultonly when the training and test data are not shuffled. Thus, we consider this result to be invalid.
《ENSEMBLE OF GENERATIVE ANDDISCRIMINATIVETECHNIQUES FORSENTIMENTANALYSIS OFMOVIEREVIEWS》文章地址：http://arxiv.org/pdf/1412.5335v6.pdf
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
