Deep Representation Learning for Sarcasm Detection in Twitte
发布时间:2023-01-15 21:38
情感分析是自然语言处理中一个非常活跃的研究领域。近来,互联网上出现了许多包含用户评论的文本资源:互联网用户的想法、论坛、社交网络、消费者调查等。考虑到数据的丰富性,自动综合多个观点对于获得对即定主题情感的概述变得至关重要。该研究对于希望了解客户对其产品的反馈意见的公司,以及希望查询关于产品或旅行的评论的人来说都很有吸引力。在过去的十年里,推特已经变得很流行,并且成为许多人日常生活的一部分。在该论文中,笔者研究了Twitter消息中的讽刺检测。尽管大多数关于讽刺检测的研究都强调词汇、句法或语用特征的使用。这些特征通常通过比喻手段来表达,如单词、表情符号和感叹号。在本文中,作者将注意力机制与深度神经模型相结合,并将其与目前最先进的特征工程方法进行了比较,探索深度学习在讽刺检测任务中的应用。作者还建立了一个关于讽刺存在的tweet手动注释数据集。因为递归神经网络(RNN)模型通常不能囊括其最终的所有重要信息隐藏状态,作者重点关注机制的影响,首先通过与长期短期记忆(LSTM)结合探究,然后与双向长期短期记忆(BLSTM)结合探究找出句子中的每个单词的相对作用。结果表明,基于注意机制下的LSTM...
【文章页数】:59 页
【学位级别】:硕士
【文章目录】:
摘要
Abstract
1 Introduction
1.1 Twitter Overview
1.2 Definition of Sarcasm
1.3 Why is sarcasm detection interesting?
1.4 Objectives and Contributions
1.5 Thesis Structure
2 Related Work
2.1 Previous Works on Sentiment Analysis
2.2 Previous Works on Sarcasm Detection
3 Background
3.1 General Neural Networks
3.2 Convolutional Neural Network
3.2.1 Convolution
3.2.2 Max Pooling
3.3 Recurrent Neural Networks
3.3.1 Long Short-Term Memory(LSTM)
3.4 Training
3.4.1 Cost Function
3.4.2 Gradient Descent
3.4.3 Backpropagation
3.5 Overfitting
3.6 Word Embedding
3.7 Attention Mechanism
4 Proposed Method
4.1 Data Collection
4.2 Data Preprocessing
4.3 Proposed Model Architecture
4.3.1 Input Layer
4.3.2 Embedding Layer
4.3.3 LSTM Layer
4.3.4 Attention Layer
4.3.5 Bidirectional Long Short-Term Memory(BLSTM)Layer
5 Data and Experiment Setup
5.1 Datasets
5.1.1 Collected Dataset
5.1.2 Gosh and Veale Dataset
5.2 Parameter Setting
5.2.1 Hardware and Software Details
5.2.2 Hyperparameter
6 Results and Analysis
6.1 Scoring Methods
6.1.1 Accuracy
6.1.2 F1-Score
6.2 Result on Ghosh and Veale Dataset
6.3 The Result of Collected Dataset
6.4 Results Comparison on Different Datasets
6.5 Evaluation
Conclusion and Future Work
References
Research Projects and Publications in Master Study
Acknowledgements
本文编号:3731518
【文章页数】:59 页
【学位级别】:硕士
【文章目录】:
摘要
Abstract
1 Introduction
1.1 Twitter Overview
1.2 Definition of Sarcasm
1.3 Why is sarcasm detection interesting?
1.4 Objectives and Contributions
1.5 Thesis Structure
2 Related Work
2.1 Previous Works on Sentiment Analysis
2.2 Previous Works on Sarcasm Detection
3 Background
3.1 General Neural Networks
3.2 Convolutional Neural Network
3.2.1 Convolution
3.2.2 Max Pooling
3.3 Recurrent Neural Networks
3.3.1 Long Short-Term Memory(LSTM)
3.4 Training
3.4.1 Cost Function
3.4.2 Gradient Descent
3.4.3 Backpropagation
3.5 Overfitting
3.6 Word Embedding
3.7 Attention Mechanism
4 Proposed Method
4.1 Data Collection
4.2 Data Preprocessing
4.3 Proposed Model Architecture
4.3.1 Input Layer
4.3.2 Embedding Layer
4.3.3 LSTM Layer
4.3.4 Attention Layer
4.3.5 Bidirectional Long Short-Term Memory(BLSTM)Layer
5 Data and Experiment Setup
5.1 Datasets
5.1.1 Collected Dataset
5.1.2 Gosh and Veale Dataset
5.2 Parameter Setting
5.2.1 Hardware and Software Details
5.2.2 Hyperparameter
6 Results and Analysis
6.1 Scoring Methods
6.1.1 Accuracy
6.1.2 F1-Score
6.2 Result on Ghosh and Veale Dataset
6.3 The Result of Collected Dataset
6.4 Results Comparison on Different Datasets
6.5 Evaluation
Conclusion and Future Work
References
Research Projects and Publications in Master Study
Acknowledgements
本文编号:3731518
本文链接:https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/3731518.html