融合多级特征的俄语推特文本情感分析  

Sentiment Analysis of Russian Tweets with Multi-level Features

作　　者：刘鑫[1] 祁瑞华[1] 徐琳宏[1] 陈恒[1] LIU Xin;QI Rui-hua;XU Lin-hong;CHEN Heng(Language Intelligence Research Center,Dalian University of Foreign Languages,Dalian 116044,China)

机构地区：[1]大连外国语大学语言智能研究中心,辽宁大连116044

出　　处：《小型微型计算机系统》2021年第6期1176-1183,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61806038)资助;教育部人文社科青年基金项目(18YJCZH208)资助;国家社会科学基金一般项目(15BYY028)资助;辽宁省教育厅科学研究经费项目(2020JYT17)资助;大连外国语大学科研基金项目(2016XJJS56)资助

摘　　要：社交媒体中俄语情感信息的深入挖掘和分析,对国家制定政治、经贸和外交战略具有重要参考价值.本文针对俄语社交媒体文本口语化、不规范、形态多样等特点,提出融合俄语形态、俚语等词级特征和特殊符号、英译情感信息等句级特征的多级特征表示方法,建立基于自注意力机制的俄语情感分类深度学习模型.针对俄语推特文本的情感分类实验表明,本文提出的多级特征能有效提升多种模型分类的F1_macro和准确率,与已有研究相比,本文模型不仅提升效果最明显,而且针对俄语情感的特征提取和分类能力更强.The in-depth mining and analysis of Russian emotion information in social media has important reference value for the formulation of national political,economic,trade and diplomatic strategies.Given that Russian social media texts are colloquial,non-standard and of diverse forms,this paper proposes a multilevel-feature representation method that combines word-level features such as Russian morphology and slang with sentence-level features such as special symbols and emotion information in English translation,and establishes a deep learning model of Russian sentiment classification based on self-attention mechanism.The sentiment classification experiment for Russian tweets shows that the multi-level features proposed in this paper can effectively improve the F1_macro and accuracy of many models in classification.Compared with the existing studies,the model in this paper not only significantly improves the effect but also has stronger abilities in features extraction and classification for Russian emotion.

关 键 词：情感分析 俄语 特征融合 自注意力机制 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]