基于双语信息和神经网络模型的情绪分类方法

摘要

文本情绪分类是自然语言处理研究中的一项基本任务.目前,已有的文本情绪分类研究大都在单语语料上进行,存在已标注样本不足、分类文本较短、信息量少等问题.为了解决上述问题,提出了一种基于双语信息和神经网络模型的情绪分类方法.首先,利用机器翻译工具对源语料进行翻译得到翻译语料;其次,将对应语言的语料进行合并,得到两组不同语言的语料;最后,将文本分别使用源语言和翻译语言进行特征表示,建立双通道长短期记忆(long short-term memory,LSTM)网络模型融合两组特征,并构建情绪分类器.实验结果表明该方法能够稳定提升文本情绪分类的性能.%Emotion classification was a fundamental learning task in natural language processing. In all existing studies, emotion classification was performed using monolingual information. Due to the challerge of scarity challenge and the limited information in short text, an approach to emotion classification based on bilingual information was proposed. Specially, a machine translation tool was adopted for both Chinese-to-English and English-to-Chinese translation. Then, these corpora were merged according to the lan-guage. Finally, a two-channel LSTM model was used to leverage both Chinese and English features. The effectiveness of the proposed approach to emotion classification was demonstrated by empirical studies.