近幾年來，由於網路社群、微部落格的普及，研究者們透過分析研究大量的社群資料，提出了多種情緒分析技術。人們的發文內容及行為會描繪出他們自身的情緒，而這些含有情緒資訊的發文可以進一步進行產品分析或其他相關應用。然而，由於人們發文習慣的不一致。導致資料蒐集時，各類情緒的資料分布不均勻。網路社群、微部落格上的情緒偵測也成為了非常困難的議題。我們提出的方法著重在彌平情緒資料的不一致，自動找出調整的權重比率調整該資料，讓訓練結果更加精確。

Recently, researches whose emphasizes Sentimental Analysis and Emotion Detections mainly go through large social data networks. People's posts behavior and their daily tweets in those networks portray emotions with valuable data, that can lead into product reviews, or emotion classification analysis. However, emotion detections on social network are sensitive, because they rely too much on the distribution of the emotion data. Some will be common because of trending topics, products, or life events, and some shows less and rare when it is unpopular or they are not trending. This results a gap in unbalanced distribution of emotions. Our study focuses on creating multiple functions for different objectives for weights learning to deal with unbalanced data. The design functions helps emotion detection systems to provide more detailed results, not just in their accuracy, but also precisions. Functions designed provides further handling improvements in different situations whose caused by unbalanced data.

摘要 i
Abstract ii
Acknowledgement iii
List of Tables vii
List of Figures viii
1 Introduction 1
2 Related Works 4
3 Methodology 7
3.1 Emotion Detection Scoring Schema . . . . . . . . . . . . . . . . . . . . . 7
3.2 Problem Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2.1 Emotional Scoring are Naive . . . . . . . . . . . . . . . . . . . . . 9
3.2.2 Emotion Detection Accuracies are Conditionally Passive . . . . . . 10
3.2.3 People’s Posts Behavior May Cause Unbalanced Datasets . . . . . 12
3.2.4 Rare and Common Emotions Exists . . . . . . . . . . . . . . . . . 12
3.3 Emotion Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Learning Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.5 Genetic Algorithm Approach . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.5.1 Chromosome Initialization . . . . . . . . . . . . . . . . . . . . . . 14
3.6 Functions Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.6.1 Accuracy-Based Approaches . . . . . . . . . . . . . . . . . . . . . 17
3.6.2 Precision-Based Approaches . . . . . . . . . . . . . . . . . . . . . 19
4 Experiment 26
4.1 Experimental Setups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.1 Training Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.1.2 Testing Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.1.3 Genetic Algorithm Base Setup . . . . . . . . . . . . . . . . . . . . 29
4.1.4 Genetic Algorithm Weight Learning Setup . . . . . . . . . . . . . 30
4.2 Experiments Baseline - Standard Emotional Degree . . . . . . . . . . . . . 31
4.3 Accuracy-Based Approach Results . . . . . . . . . . . . . . . . . . . . . . 31
4.3.1 Function 1: Naive Weighted Accuracy Optimization . . . . . . . . 32
4.3.2 Function 2: Precision Supported Accuracy Optimization . . . . . . 34
4.4 Precision-Based Approach Results . . . . . . . . . . . . . . . . . . . . . . 36
4.4.1 Function 3: Overall Precision Evaluation . . . . . . . . . . . . . . 37
4.4.2 Performance Comparison - Accuracy and Precision Approaches . . 40
4.4.3 Rare and Common Emotions Trend . . . . . . . . . . . . . . . . . 41
4.4.4 Focused Rare and Common Emotions Evaluation(Function 4) . . . 42
4.5 Overall Experiment Comparisons . . . . . . . . . . . . . . . . . . . . . . . 45
4.5.1 Common Emotions Overall Comparison . . . . . . . . . . . . . . . 46
4.5.2 Rare Emotions Overall Comparison . . . . . . . . . . . . . . . . . 48
4.6 Functions Evaluation Comparison . . . . . . . . . . . . . . . . . . . . . . 49
4.6.1 First Function Evaluation . . . . . . . . . . . . . . . . . . . . . . . 49
4.6.2 Second Function Evaluation . . . . . . . . . . . . . . . . . . . . . 49
4.6.3 Third Function Evaluation . . . . . . . . . . . . . . . . . . . . . . 50
4.6.4 Fourth Function Evaluation . . . . . . . . . . . . . . . . . . . . . 50
5 Conclusion and Future Work 51
5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.2 Future Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
References 53

[1] Andrew J. McMinn, Daniel Tsvetkov, Tsvetan Yordanov, Andrew Patterson, Rrobi
Szk, Jesus A. Rodriguez Perez, and Joemon M. Jose. An interactive interface for
visualizing events on twitter. In Proceedings of the 37th International ACM SIGIR
Conference on Research & Development in Information Retrieval, SIGIR
’14, pages 1271–1272, New York, NY, USA, 2014. ACM. ISBN 978-1-4503-2257-
7. doi: 10.1145/2600428.2611189. URL http://doi.acm.org/10.1145/
2600428.2611189.
[2] Bo Pang, Lillian Lee, and Shivakumar Vaithyanathan. Thumbs up?: sentiment classification
using machine learning techniques. In Proceedings of the ACL-02 conference
on Empirical methods in natural language processing-Volume 10, pages 79–86.
Association for Computational Linguistics, 2002. URL http://dl.acm.org/
citation.cfm?id=1118704.
[3] Minqing Hu and Bing Liu. Mining and summarizing customer reviews. In Proceedings
of the tenth ACM SIGKDD international conference on Knowledge discovery
and data mining, pages 168–177. ACM, 2004.
[4] Carlos Argueta, Yi-Shin. Chen, and Elvis Saravia. Unsupervised graph-based patterns extraction for emotion classification. In 2015 IEEE/ACM International Conference
on Advances in Social Networks Analysis and Mining (ASONAM 2015), 2015
[5] Robert Plutchik. The natural of emotions. American Scientist, 89(4):344–350, July
2001.
[6] Wenbo Wang, Lu Chen, Krishnaprasad Thirunarayan, and Amit P Sheth. Harnessing
twitter ”big data” for automatic emotion identification. In Privacy, Security, Risk and
Trust (PASSAT), 2012 International Conference on and 2012 International Conference
on Social Computing (SocialCom), pages 587–592. IEEE, 2012.
[7] S. Wen and X. Wan. Emotion Classification in Microblog Texts Using Class
Sequential Rules. In Twenty-Eighth AAAI Conference on Artificial Intelligence,
2014. URL http://www.aaai.org/ocs/index.php/AAAI/AAAI14/
paper/view/8209.
[8] Ahmed Esmin, RL de Oliveira, Stan Matwin, et al. Hierarchical classification approach
to emotion recognition in twitter. In Machine Learning and Applications
(ICMLA), 2012 11th International Conference on, volume 2, pages 381–385. IEEE,
2012.
[9] Jin-Ji Li Jungi Kim and Jong-Hyeok Lee. Discovering the discriminative views: Measuring
term weights for sentiment analysis. 2009.
[10] Nikolaos Pappaos and Andrei Popescu. Explaining the stars: Weighted multipleinstance
learning for aspect-based sentiment analysis. 2014