用于微博情感分析的一种情感语义增强的深度学习模型

发布时间：2018-02-08 20:10

本文关键词： 微博情感分析深度学习卷积神经网络自然语言处理社交网络　出处：《计算机学报》2017年04期 　论文类型：期刊论文

【摘要】：基于神经语言模型的词向量表示技术能够从大规模的未标注文本数据集中自动学习词语的有效特征表示,已经在许多自然语言处理任务及研究中取得重要进展.微博中的表情符号是微博情感分析最重要的特征之一,已有大量的研究工作在探索有效地利用表情符号来提升微博情感分类效果.借助词向量表示技术,为常用表情符号构建情感空间的特征表示矩阵R~E;基于向量的语义合成计算原理,通过矩阵R~E与词向量的乘积运算完成词义到情感空间的映射;接着输入到一个MCNN(Multi-channel Convolution Neural Network)模型,学习一个微博的情感分类器.整个模型称为EMCNN(Emotion-semantics enhanced MCNN),将基于表情符号的情感空间映射与深度学习模型MCNN结合,有效增强了MCNN捕捉情感语义的能力.EMCNN模型在NLPCC微博情感评测数据集上的多个情感分类实验中取得最佳分类性能,并在所有性能指标上超过目前已知文献中的最好分类效果.在取得以上分类性能提升的同时,EMCNN相对MCNN的训练耗时在主客观分类时减少了36.15%,在情感7分类时减少了33.82%.
[Abstract]:Word vector representation technology based on neural language model can automatically learn effective feature representation of words from large scale unlabeled text dataset. Important progress has been made in many natural language processing tasks and studies. The emoji in Weibo is one of the most important features of Weibo's affective analysis. A great deal of research has been done to explore the effective use of emoji to improve Weibo's affective classification. The feature representation matrix of affective space is constructed for common emoji. Based on the computation principle of semantic synthesis of vector, the mapping of word meaning to emotion space is completed by multiplying the product of matrix Rae and word vector, and then an MCNN(Multi-channel Convolution Neural network model is inputted. Learn a Weibo's affective classifier. The whole model, called EMCNN(Emotion-semantics enhanced MCNNN, combines emoji based affective space mapping with depth learning model MCNN. It effectively enhances the ability of MCNN to capture emotional semantics. EMCNN model achieves the best classification performance in a number of affective classification experiments on the emotional evaluation dataset of NLPCC Weibo. At the same time, the training time of MCNN relative to MCNN is reduced by 36.15 in subjective and objective classification, and 33.82 in emotional 7 classification.
【作者单位】：武汉大学计算机学院;武汉大学软件工程国家重点实验室;
【基金】：国家自然科学基金(61303115,61373039,61472290) 高等学校博士学科点专项科研基金(2013014111002512)资助~~
【分类号】：TP391.1;TP18

【相似文献】