预训练模型下融合注意力机制的多语言文本情感分析方法  

Multilingual Text Emotional Analysis with Pre-trained Model and Attention Mechanism

作　　者：胡德敏[1] 褚成伟 胡晨 胡钰媛 HU De-ming;CHU Cheng-wei;HU Chen;HU Yu-yuan(School of Optical-Electrical&Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《小型微型计算机系统》2020年第2期278-284,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61170277,61472256)资助;上海市教委科研创新重点项目(12zz137)资助;上海市一流学科建设项目(S1201YLXK)资助.

摘　　要：多语言文本的情感分析是情感分析领域的重要问题之一,而现有的情感分析方法着重于对单语言文本的研究.本文针对中英混合文本提出了一种细粒度情感分析模型,通过基于大规模语料的预训练语言模型得到上下文语义相关的词向量,将词向量输入双向LSTM网络学习文本的情感表征,使用多语言注意力机制分别针对单语和双语文本提取关键情感表征,最终通过并行融合的方式提升情感分类效果.本文使用NLPCC2018多语言文本情绪分析任务数据集进行细粒度情感分析,对比评测任务中的最好结果,本文模型得到的宏平均F1值提高至0. 581,表明了本文方法的有效性.The multilingual text emotional analysis is an important issue in the field of emotional analysis. However,the existing researches of emotional analysis mostly focus on monolingual text. This paper proposed a fine-grained emotional analysis model for ChineseEnglish mixed texts. Firstly,the context-dependent word vectors were given by pre-trained model based on large-scale corpus to avoid the possible ambiguity in the text. Then,we trained the state vectors of the text by BiLSTM model and divided them into two parts:monolingual states and multilingual states. We applied the attention mechanism on those three output states separately to get the final vectors. Finally,we integrated those vectors to predict the emotion. We used the datasets produced by NLPCC 2018 multilingual emotional analysis task. The experimental results showed that our model outperformed the best model of the multilingual emotional analysis task with the Macro-F1 value increased to. 581.

关 键 词：情感分析 多语言文本 预训练 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]