基于双向长短时记忆单元和卷积神经网络的多语种文本分类方法

针对日渐丰富的多语种文本数据，为了实现对同一类别体系下不同语种的文本分类，充分发挥多语种文本信息的价值，提出一种结合双向长短时记忆单元和卷积神经网络的多语种文本分类模型BiLSTM-CNN模型。针对每个语种，利用双向长短时记忆神经网络提取文本特征，并引入卷积神经网络进行特征优化，获得各语种更深层次的文本表示，最后将各语种的文本表示级联输入到softmax函数预测类别。在中英朝科技文献平行数据集上进行了实验验证，实验结果表明，该方法相比于基准方法分类正确率提高了4%，且对任一语种文本均能正确分类，具有良好的扩展性。

Multilingual text classification method based on bi-directional long short-term memory and convolutional neural network

In order to realize the text classification of different languages in the same category system and make full use of the value of multilingual text information, this paper proposed a multilanguage text classification model BiLSTM-CNN which combined bidirectional long short-term memory and convolutional neural networks. For each language, it extracted the text features through the two-way long-term memory neural network, and introduced the convolutional neural network to extract the text local information for feature optimization, so as to realize the distributed text representation of different language documents. Finally, it cascaded the text representation of each language into the softmax function prediction category. Experiments on parallel datasets of Chinese, British and Korean scientific and technological documents show that the proposed multilingual text classification model has a 4% improvement over the benchmark method, and can correctly classify any linguistic text with good expansibility.