[1]高 慧,荀亚玲*,王林青.基于多通道融合特征网络的文本情感分析[J].计算机技术与发展,2023,33(11):175-181.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 026]
　GAO Hui,XUN Ya-ling*,WANG Lin-qing.Text Sentiment Analysis Based on Multi-channel Fusion Feature Network[J].,2023,33(11):175-181.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 026]

点击复制

基于多通道融合特征网络的文本情感分析()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年11期

页码:

175-181

栏目:

人工智能

出版日期:

2023-11-10

文章信息/Info

Title:

Text Sentiment Analysis Based on Multi-channel Fusion Feature Network

文章编号:

1673-629X(2023)11-0175-07

作者:

高 慧; 荀亚玲* ; 王林青

太原科技大学 计算机科学与技术学院,山西 太原 030024

Author(s):

GAO Hui; XUN Ya-ling* ; WANG Lin-qing

School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China

关键词:

情感分析; ChineseBERT; 多通道融合特征; 内置注意力简单循环单元; 软注意力

Keywords:

sentiment analysis; ChineseBERT; multi-channel fusion features; built-in attention SRU; soft attention

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 11. 026

摘要:

针对现有文本情感分析基础深度学习模块特征提取不够全面,语义表示不准确及训练效率低等问题,提出了基于多通道融合特征网络的文本情感分析模型。 首先,采用针对汉字优化的预训练模型 ChineseBERT 提取文本的动态词向量表征,解决静态词向量存在的无法表示多义词问题,提升词向量语义表征质量;然后,通过多通道融合特征网络全面捕捉文本不同尺度下的语义特征融合向量表示,增强模型对文本深层次情感特征的学习能力;并利用软注意力机制计算每个特征对情感极性类型识别的影响权重,赋予关键特征更高权重,避免无关特征对结果造成干扰;最后,由线性层输出文本情感分类结果。 在 SMP2020 微博疫情相关情绪分类评测数据集、购物评论数据集和酒店评论数据集上进行实验验证,分别取得了 76. 59% 、97. 59% 和 95. 72% 的 F1 分数以及 76. 6% 、97. 59% 和 95. 73% 的准确率,高于近期表现优秀的对比深度学习模型,验证了该模型在文本情感分析任务上的有效性。

Abstract:

To address the problems of incomplete feature extraction,inaccurate semantic representation and low training efficiency of the　existing text sentiment analysis basic deep learning module, a text sentiment analysis model based on multi - channel fusion featurenetwork is proposed. First of all, the dynamic word vector representation of text is extracted by using the pre training modelChineseBERT optimized for Chinese characters to solve the problem that static word vectors cannot represent polysemy words, andimprove the semantic representation quality of word vectors; then,the multi channel fusion feature network is used to capture the semanticfeature fusion vector representation at different scales of the text,so as to enhance the learning ability of the model to the deep emotionalfeatures of the text; the soft attention mechanism is used to calculate the influence weight of each feature on the recognition of emotionalpolarity type,and the key features are given higher weight to avoid the interference of irrelevant features on the results; finally,the text emotion classification results are output from the linear layer. The experimental verification was carried out on the SMP2020 microblog epidemic related emotion classification and evaluation data set,shopping review data set and hotel review data set,and the F1 scores of　76. 59% ,97. 59% and 95. 72% and the accuracy of 　

76. 6% ,97. 59% and 95. 73% were obtained respectively,which was higher thanthat of the contrast in-depth learning model with excellent performance in recent years,and verified the effectiveness of the model in textemotion analysis tasks.

常用功能