【摘要】 诗词的风格分类其实质是文本分类问题。相比于传统的基于词袋或N-Gram的文本分类方法忽略字词之间的上下文联系,循环神经网络将文本分类处理为序列化分类问题,取得更高的效率和准确率。将双向循环神经网络(BiLSTM)运用到中文古诗词风格分类中,通过诗词字、词向量与BiLSTM结合构建基于深度神经网络的中文古诗词分类模型,与传统的分类模型相比,显著的提高分类效果。并通过实验验证基于中文古诗词语料库的诗词字或词向量模型比基于现代汉语语料库的词向量模型分类准确率更高。更多还原

【Abstract】 The essence of poetry-style classification is text classification. Unlike traditional word-bag-based or N-Gram-based textclassification methods that ignore the contextual relationship between words, a recurrent neural network treats text classification as serialization classification, achieving higher efficiency and accuracy. Applies a Bidirectional Recurrent Neural Network(BiLSTM)to the classification of Chinese ancient poetry styles, and constructs a Chinese ancient poetry-classification model based on a deep neural network by combining poetry character embedding and word embedding with BiLSTM. Compared with traditional classification models, the classification effect is significantly improved. Verifies through experiments that the poetry character-or word-embedding model based on the Chinese ancient poetry corpus has higher classification accuracy than the word-embedding model based on the modern Chinese corpus.更多还原