Character-level Convolutional Network for Text Classification Applied to Chinese Corpus

Compared with word-level and sentence-level convolutional neural networks (ConvNets), the character-level ConvNets has a better applicability for misspellings and typos input. Due to this, recent researches for text classification mainly focus on character-level ConvNets. However, while the majority of these researches employ English corpus for the character-level text classification, few researches have been done using Chinese corpus. This research hopes to bridge this gap, exploring character-level ConvNets for Chinese corpus test classification. We have constructed a large-scale Chinese dataset, and the result shows that character-level ConvNets works better on Chinese character dataset than its corresponding pinyin format dataset, which is the general solution in previous researches. This is the first time that character-level ConvNets has been applied to Chinese character dataset for text classification problem.