| 刘文远,郭智存,郭丁丁.旅游场景下的基于深度学习的文本方面级细粒度情感分类[J].高技术通讯(中文),2022,32(1):22~32 |
| 旅游场景下的基于深度学习的文本方面级细粒度情感分类 |
| Text aspect-level fine-grained sentiment classification based on deep learning in travel scenarios |
| |
| DOI：10.3772/j.issn.1002-0470.2022.01.003 |
| 中文关键词: 情感分类； 深度学习； 神经网络； 注意力机制 |
| 英文关键词: sentiment classification, deep learning, neural network, attention |
| 基金项目: |
|
| 摘要点击次数: 1523 |
| 全文下载次数: 1035 |
| 中文摘要: |
| 方面级细粒度情感分类是指针对文本数据，分析其在指定方面的情感极性。由于获取到的评论样本往往涉及不同的方面，导致各个方面的情感极性不平衡。为了减少不平衡数据对模型训练的影响，本文提出了一种新的数据平衡方法——批处理平衡方法（BB），用来平衡多标签多类别数据。同时，由于评论文本蕴含多个方面，传统模型结构往往每次只能预测一个方面的情感。为了提高情感挖掘效率，本文提出了自动关注不同方面的情感注意力网络——双向循环卷积注意力网络（Attn-Bi-LCNN）模型。模型会同时关注不同方面的不同情感信息形成情感语义矩阵，根据情感矩阵进行情感预测。对比实验表明，模型取得了更好的预测结果和更快的运算速度。 |
| 英文摘要: |
| Aspect-level fine-grained sentiment classification is to analyze the emotional polarity of the text data in a given aspect. As the sample of comments often involves different aspects, the emotional polarity of each aspect is unbalanced. In order to reduce the impact of unbalanced data on model training, this paper proposes a new data balance method——batch balance (BB), which is used to balance multi-label and multi-category data. At the same time, because the commentary text itself contains multiple aspects, the traditional model structure can only predict one aspect of emotion at a time. In order to improve the efficiency of emotion mining, this paper proposes an attention network that automatically focuses on different aspects of emotions——a bidirectional circular convolutional attention network model. The model will pay attention to different affective information from different aspects at the same time to form the affective semantic matrix, and make affective prediction according to the affective matrix. The comparative experiments show that the proposed model achieves better prediction results and faster computation speed. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
