一种基于联合深度学习模型的情感分类方法  

An emotional classification method based on joint deep learning model

作　　者：杨艳[1] 徐冰[1] 杨沐昀[1] 赵晶晶 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院机器智能与翻译实验室,黑龙江哈尔滨150001

出　　处：《山东大学学报（理学版）》2017年第9期19-25,共7页Journal of Shandong University(Natural Science)

基　　金：国家自然科学基金资助项目(61402134)

摘　　要：针对情感分类问题中长句和短句具有不同的建模特点,提出了一种基于联合深度学习模型的情感分类方法。该方法融合长短期记忆(long-short term memory,LSTM)模型与卷积神经网络(convolutional neural network,CNN)对影视评论数据进行情感极性判别;采用LSTM对上下文进行建模,通过逐词迭代得到上下文的特征向量;采用CNN模型从词向量序列中自动发现特征,抽取局部特征并整合成全局特征来提高分类效果。所提出的方法在COAE2016评测的任务2的情感极性分类任务中,取得最高的系统准确率。According to the problems of emotional classification in the modeling of long and short sentences with differ- ent characteristics, this paper proposed a classification algorithm based on the model of joint deep learning. Long-short term memory （LSTM） model and convolutional neural network （CNN） were combined to discriminate the emotional polarity of film reviews. LSTM model was used to model context, word iteration was used to get feature vector context, and CNN model was used to automatically discover features from the word vector sequence. Local features was extrac- ted and integrated into global features to improve classification results, The proposed method had the highest system ac- curacy in COAE2016 evaluation task 2.

关 键 词：情感分类 长短期记忆模型 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]