基于多跳动态记忆网络和情感词典的情感分析模型

沈金金; 陈荔

文章摘要

沈金金,陈荔.基于多跳动态记忆网络和情感词典的情感分析模型[J].情报工程,2022,8(2):003-018

基于多跳动态记忆网络和情感词典的情感分析模型

Emotion Analysis Model Based on Dynamic Memory Network and Emotion Dictionary

DOI：10.3772/j.issn.2095-915X.2022.02.001

中文关键词: 动态记忆网络；情感词典；注意力机制；情感分析

英文关键词: Dynamic memory network; sentiment dictionary; attention mechanism; sentiment analysis

基金项目:国家自然科学基金项目“产业互联”智造“供需网的结构、演化及其动力学研究”（71871144）。

作者	单位
沈金金	上海理工大学上海 200093
陈荔	上海理工大学上海 200093

摘要点击次数: 1060

全文下载次数: 1194

中文摘要:

[ 目的 / 意义 ] 文本情感分析是自然语言处理的一大重要分支。论文结合了深度学习模型在特征提取方面的优势以及情感词典对网络词情感识别的敏感性，提出了一种将动态记忆网络和情感词典方法相结合的网络文本情感分类模型。[ 方法 / 过程 ] 在传统的动态记忆网络中设计情感问题向量，利用基于注意力机制的多跳结构识别并提取句子情感特征，同时构建扩充网络情感词的情感词典并将情感分值与多跳记忆网络所得情感分类结果进行线性加权，减少了对于情感词库的完备性、判断规则质量以及事先标注语料的依赖，提升了训练效率。[ 结果 / 结论 ] 在 WEIBO_SENTI_100K和 NLPCC2013 两个数据集上进行实验，证明与单个模型相比，论文模型在两个数据集上都获得了更好的分类效果。

英文摘要:

[Objective/Significance] Text sentiment analysis is an important branch of natural language processing. Combining the advantages of deep learning model in feature extraction and the sensitivity of sentiment dictionary to identify the emotions of emerging online words, this paper proposes an online text emotion classification model combining dynamic memory network and sentiment dictionary. [Methods/Process] Emotion problem vectors are designed in the traditional dynamic memory network,and the multi-hop structure based on attention mechanism is used to identify and extract sentence emotion features.Meanwhile, the sentiment dictionary with extended network emotion words is constructed, and the emotion score is weighted with the emotion classification results obtained from multi-hop memory network. It reduces the dependence on the completeness of sentiment lexicon, the quality of judgment rules and the pre-annotation corpus, and improves the training efficiency.[Results/Conclusions]Experiments on two datasets, WEIBO_SENTI_100K and NLPCC2013, prove that the proposed model achieves better classification effect on both datasets compared with a single model.

查看全文查看/发表评论下载PDF阅读器

关闭