融合注意力机制和CNN-GRNN模型的读者情绪预测  

Attention-based convolutional-gated recurrent neural network for reader's emotion prediction

作　　者：张琦 彭志平 ZHANG Qi;PENG Zhiping(School of Computers,Guangdong University of Technology,Guangzhou 510006,China;School of Computer and Electronic Information,Guangdong University of Petrochemical Technology,Maoming,Guangdong 525000,China)

机构地区：[1]广东工业大学计算机学院,广州510006 [2]广东石油化工学院计算机与电子信息学院,广东茂名525000

出　　处：《计算机工程与应用》2018年第13期168-174,共7页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61272382;No.61672174);广东省科技计划项目(No.2015B020233019)

摘　　要：针对主流面向文本的读者情绪预测算法难以捕捉文本中复杂的语义和语法信息,以及局限于使用多标签分类方法的问题,提出一种融合注意力机制和卷积门限循环神经网络的读者情绪预测方法。该方法将文本划分为多个句子,利用卷积神经网络从每个句子中提取不同粒度的n-gram信息,构建句子级别的特征表示;然后通过门限循环神经网络顺序地集成这些句子特征,并利用注意力机制自适应地感知上下文信息提取影响读者情绪的文本特征;最后利用softmax回归进行细粒度的读者情绪分布预测。在雅虎新闻读者情感分析数据集上的实验结果证明了该方法的有效性。The past reader＇s emotion prediction methods are unable to capture the complex semantic and grammatical information in the document, and have mostly used in multi-label classification technology, which limit its development and application. To solve this problem, an improved method named attention-based convolutional-gated recurrent neural network for reader＇s emotion prediction is presented. This method firstly divides the document into several sentences, and then adopts convolutional neural network to produce sentence-level representations from word vectors. Such useful sentence features can be sequentially integrated using gated recurrent neural network, and a novel attention mechanism is proposed to build a document-level representation according to their contribution to reader＇s emotion prediction. Finally, a softmax regression is applied to predict reader＇s emotion distributions. Experimental results on Yahoo news corpus demonstrate that the proposed method achieves better accuracy compared with state-of-the-art methods.

关 键 词：情感分析 读者情绪预测 卷积神经网络 门限循环神经网络 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]