Please wait a minute...
Vis Inf
Vis Inf  2020, Vol. 4 Issue (2): 72-85    DOI: 10.1016/j.visinf.2020.04.006
论文     
面向ConvNet的词语级情感分解：情感分类器的可视化
Piyush Chawla, Subhashis Hazarika, Han-Wei Shen
The Ohio State University, Columbus, 43210, USA
Token-wise sentiment decomposition for ConvNet: Visualizing a sentiment classifier
Piyush Chawla, Subhashis Hazarika, Han-Wei Shen
The Ohio State University, Columbus, 43210, USA
 全文: PDF 
摘要： 卷积神经网络是自然语言处理和人工智能中最重要且应用最为广泛的一种构造。在许多应用中，它们已经达到了当前最优的性能，且所需的训练时间比其他方法少。但是，由于其可解读性有限，与基于注意力的模型（RNN和自我关注模型）相比，它们较少受实践者的青睐，而后者可以通过分析注意力权重热图来做更为直观地解读。 本文提出了一种可视化技术，可用于理解基于文本的CNN模型的内部工作过程。同时还展示了如何使用此方法生成对抗性示例并找到训练数据的不足之处。
关键词： 卷积神经网络;  可视化;  情感分析    
Abstract: Convolutional neural networks are one of the most important and widely used constructs in natural language processing and AI in general. In many applications, they have achieved state-of-the-art performance, with training time faster than the other alternatives. However, due to their limited interpretability, they are less favored by practitioners over attention-based models, like RNNs and self-attention (Transformers), which can be visualized and interpreted more intuitively by analyzing the attention-weight heat-maps. In this work, we present a visualization technique that can be used to understand the inner workings of text-based CNN models. We also show how this method can be used to generate adversarial examples and learn the shortcomings of the training data.
Key words: Convolutional neural networks    Visualization    Sentiment analysis
出版日期: 2020-06-02
通讯作者: Piyush Chawla     E-mail: chawla.81@osu.edu
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章　　
Piyush Chawla
Subhashis Hazarika
Han-Wei Shen

引用本文:

Piyush Chawla, Subhashis Hazarika, Han-Wei Shen. Token-wise sentiment decomposition for ConvNet: Visualizing a sentiment classifier. Vis Inf, 2020, 4(2): 72-85.

链接本文:

http://www.zjujournals.com/vi/CN/10.1016/j.visinf.2020.04.006    或    http://www.zjujournals.com/vi/CN/Y2020/V4/I2/72

[1] Xiaoyan Kui, Huihao Lv, Zhengliang Tang, Haowen Zhou, Wang Yang, Jinqiu Li, Jialin Guo, Jiazhi Xia. TVseer：一个电视收视率的可视分析系统[J]. Vis Inf, 2020, 4(3): 1-11.
[2] Sujia Zhu, GuodaoSun, Qi Jiang, Meng Zha, Ronghua Liang. 自动信息图和可视化推荐综述 [J]. Vis Inf, 2020, 4(3): 24-40.
[3] Honghui Mei, Huihua Guan, Chengye Xin, Xiao Wen, Wei Chen. DataV: 面向大尺度高分辨率显示装置的敏捷可视化设计与实现 [J]. Vis Inf, 2020, 4(3): 12-23.
[4] ChrisBryan, Aditi Mishra, Hidekazu Shidara, Kwan-Liu Ma. 不同任务背景下对配有文本修饰的叙事式可视化结果的凝视行为分析[J]. Vis Inf, 2020, 4(3): 41-50.
[5] Wenbin He, Junpeng Wang, Hanqi Guo, Han-Wei Shen, Tom Peterka. CECAV-DNN：使用深度神经网络进行集合比较和可视化 [J]. Vis Inf, 2020, 4(2): 109-121.
[6] Piyush Chawla, Subhashis Hazarika, Han-Wei Shen. 面向ConvNet的词语级情感分解：情感分类器的可视化 [J]. Vis Inf, 2020, 4(2): 132-141.
[7] Kadri Umbleja, Manabu Ichino, Hiroyuki Yaguchi. 改进符号数据可视化以促进模式识别和知识发现[J]. Vis Inf, 2020, 4(1): 23-31.
[8] Ruochen Cao, Subrata Dey, Andrew Cunningham, James Walsh, Ross T.Smith, Joanne E.Zucco, Bruce H. Thomas. 考察叙事结构在数据视频中的运用 [J]. Vis Inf, 2020, 4(1): 8-22.
[9] Lei Shi, Qi Liao, Hanghang Tong, Yifan Hu, Chaoli Wang, Chuang Lin, Weihong Qian. OnionGraph:层次拓扑与属性多变量，网络可视化 [J]. Vis Inf, 2020, 4(1): 43-57.
[10] Kecheng Lu, Chaoli Wang, Keqin Wu, Minglun Gong, Yunhai Wang. 基于块对应的时变体数据挖掘的统一框架 [J]. Vis Inf, 2019, 3(4): 157-165.
[11] Davide Ceneda, Theresia Gschwandtne, Silvia Miksch. 在交互可视化中不同的引导方式对交互绩效和心理状态的影响 [J]. Vis Inf, 2019, 3(4): 177-191.
[12] Cui Xie, Mingkui Li, Haoying wang, Junyu Dong. 海洋数据可视化分析综述 [J]. Vis Inf, 2019, 3(3): 113-128.
[13] Xueyi Chen, Liming Shen, Ziqi Sha, Richen Liu, Siming Chen, Genlin Ji, ChaoTan. 时空模拟数据可视化的多空间分析技术综述 [J]. Vis Inf, 2019, 3(3): 129-139.
[14] Huan Liu, Sichen Jin, Yuyu Yan, Yubo Tao, Hai Lin. 利用主题子轨迹对出租车轨迹进行可视分析[J]. Vis Inf, 2019, 3(3): 140-149.
[15] Junhua Lu, Xiao Xie,  Ji Lan,  Tai-Quan Peng,  Yingcai Wu,  Wei Chen. BeXplorer:大型多人在线角色扮演类游戏中玩家交流与消费行为动态关联的可视分析 [J]. Vis Inf, 2019, 3(2): 87-101.