文本情感分析是對文本中的某段已知文字進行情緒分類。互動式情感分析是一個非常具有挑戰性的領域。而有鑑於情緒在人際互動中的重要性，目前已有許多在電腦中探測人類對話情緒的實驗，但沒有一個大眾普遍都同意的情緒定義和類型，以及人們對於文字情緒的理解也都不盡相同，所以這是一個非常困難的議題。因此，本研究將以康乃爾大學所提供的Movie Dialog Corpus使用深度學習模型透過談話情境，追蹤連續對話的情緒。
本研究的目的是希望可以透過單一句子的情緒分析，經過權重的分配，預測最新一句，以驗證雙向情緒的遞延。同時，在建立模型的過程中研究類別權重、語意規則以及不同機器學習方法對預測能力的影響。
實驗發現如果有加入類別權重，在多類別分類中，預測類別不會集中於數量較多的類別。而利用語意規則，Convolutional LSTM在各類別或者整體的預測正確的狀況，皆明顯比沒有使用語意規則還要好。最後，對話組的句子經過權重相乘相加，正確預測的能力隨著資訊量越多而越強，說明了可有效遞延對話的情緒以及加強預測能力。

Text sentiment analysis is the emotion classification of a particular piece of text. Interactive sentiment analysis is a very challenging research field. Given the importance of emotions in interpersonal interactions, there were many experimental studies in which emotions were detected by computers. However, there is no commonly agreed emotion definition and emotion types, and people's comprehensions on text emotions are different. In this study, we use the deep learning model and Movie Dialog Corpus provided by Cornell University to track the emotions of continuous dialogue through context.
This study aims to verify the effect of emotion aggregation and propagation in conversations through the emotion analysis of a single sentence and the use of weights for multiple sentences. Additionally, the effects of category weights, semantic rules and different machine learning methods on classification are discussed in along with the process of building models.
The experimental results shows that in our multi-class classification problem, if class weights are used, the predicted results will not be deviated toward the classes with large numbers of data. Also, the cases of using Convolutional LSTM with semantic rules are significantly better than those without semantic rules. Finally, the emotions of the consecutive sentences of in the dialogue are weighted and accumulated in prediction. The results indicate that the emotions can be propagated through time and thus effectively enhance the classification performance.

論文審定書 i
誌謝 ii
摘要 iii
Abstract iv
第一章、緒論 1
1.1研究背景 1
1.2研究動機 1
1.3研究目的 2
1.4研究架構 3
第二章、文獻探討 4
2.1情感分析 4
2.1.1單向文本情感分析 4
2.1.2雙向文本情感分析 6
2.2 深度學習 8
2.2.1循環神經網路(Recurrent Neural Networks, RNNs) 8
2.2.2 長短期記憶(Long Short Term Memory Network, LSTM) 10
第三章、研究方法 12
3.1 資料蒐集以及格式定義 13
3.2 資料前處理 15
3.3 類別平衡 19
3.3.1 類別權重 20
3.3.2 以二元分類驗證類別平衡之效益 21
3.4 分類模型建立 22
3.5 對話 25
第四章、研究結果 27
4.1評估準則 27
4.2二元分類結果分析 29
4.3各模型比較 36
4.3.1 類別平衡的影響 39
4.3.2 語意規則的影響 42
4.3.3 同時沒有類別平衡與語意規則 45
4.4 形成對話的影響 48
4.4.1 單句對話 48
4.4.2多句對話 52
4.5 與Watson系統之比較 54
4.6 綜合討論 55
第五章、結論 56
5.1結論 56
5.2未來研究建議 57
參考文獻 58

Mostafa, M. M. (2013). More than words: Social networks’ text mining for consumer brand sentiments. Expert Systems with Applications, 40(10), 4241-4251.
Khan, F. H., Bashir, S., & Qamar, U. (2014). TOM: Twitter opinion mining framework using hybrid classification scheme. Decision Support Systems, 57, 245-257.
Wang, B., Liakata, M., Zubiaga, A., Procter, R., & Jensen, E. (2016). SMILE: Twitter Emotion Classification using Domain Adaptation. In 25th International Joint Conference on Artificial Intelligence (p. 15).
Desmet, B., & Hoste, V. (2013). Emotion detection in suicide notes. Expert Systems with Applications, 40(16), 6351-6358.
Vanzo, A., Croce, D., & Basili, R. (2014). A context-based model for sentiment analysis in twitter. In Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers (pp. 2345-2354).
Yan, J. L. S., & Turtle, H. R. (2016). Exposing a Set of Fine-Grained Emotion Categories from Tweets. In 25th International Joint Conference on Artificial Intelligence (p. 8).
Wang, L., & Cardie, C. (2016). A piece of my mind: A sentiment analysis approach for online dispute detection. arXiv preprint arXiv:1606.05704.
Zhang, L., & Chen, C. (2016, December). Sentiment classification with convolutional neural networks: an experimental study on a large-scale Chinese conversation corpus. In Computational Intelligence and Security (CIS), 2016 12th International Conference on (pp. 165-169). IEEE.
Nguyen, H., & Nguyen, M. L. (2017, August). A Deep Neural Architecture for Sentence-Level Sentiment Classification in Twitter Social Networking. In International Conference of the Pacific Association for Computational Linguistics (pp. 15-27). Springer, Singapore.
Phan, D. A., Shindo, H., & Matsumoto, Y. (2016). Multiple emotions detection in conversation transcripts. In Proceedings of the 30th Pacific Asia Conference on Language, Information and Computation: Oral Papers (pp. 85-94).
Gupta, U., Chatterjee, A., Srikanth, R., & Agrawal, P. (2017). A sentiment-and-semantics-based approach for emotion detection in textual conversations. arXiv preprint arXiv:1707.06996.
Huang, C. Y., Labetoulle, T., Huang, T. H. K., Chen, Y. P., Chen, H. C., Srivastava, V., & Ku, L. W. (2017). MoodSwipe: A Soft Keyboard that Suggests Messages Based on User-Specified Emotions. arXiv preprint arXiv:1707.07191.
Appel, O., Chiclana, F., Carter, J., & Fujita, H. (2016). A hybrid approach to the sentiment analysis problem at the sentence level. Knowledge-Based Systems, 108, 110-124.
Tang, D., Wei, F., Yang, N., Zhou, M., Liu, T., & Qin, B. (2014). Learning sentiment-specific word embedding for twitter sentiment classification. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (Vol. 1, pp. 1555-1565).
Bifet, A., & Frank, E. (2010, October). Sentiment knowledge discovery in twitter streaming data. In International conference on discovery science (pp. 1-15). Springer, Berlin, Heidelberg.
Pagolu, V. S., Reddy, K. N., Panda, G., & Majhi, B. (2016, October). Sentiment analysis of Twitter data for predicting stock market movements. In Signal Processing, Communication, Power and Embedded System (SCOPES), 2016 International Conference on (pp. 1345-1350). IEEE.
Wang, Y., Huang, M., & Zhao, L. (2016). Attention-based lstm for aspect-level sentiment classification. In Proceedings of the 2016 conference on empirical methods in natural language processing (pp. 606-615).
Kiritchenko, S., Zhu, X., & Mohammad, S. M. (2014). Sentiment analysis of short informal texts. Journal of Artificial Intelligence Research, 50, 723-762.
Bothe, C., Magg, S., Weber, C., & Wermter, S. (2017, September). Dialogue-Based Neural Learning to Estimate the Sentiment of a Next Upcoming Utterance. In International Conference on Artificial Neural Networks (pp. 477-485). Springer, Cham.
Huang, M., Cao, Y., & Dong, C. (2016). Modeling rich contexts for sentiment classification with lstm. arXiv preprint arXiv:1605.01478.
Murray, G. (2016, May). Uncovering hidden sentiment in meetings. In Canadian Conference on Artificial Intelligence (pp. 64-72). Springer, Cham.
Herzig, J., Feigenblat, G., Shmueli-Scheuer, M., Konopnicki, D., Rafaeli, A., Altman, D., & Spivak, D. (2016). Classifying emotions in customer support dialogues in social media. In Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue (pp. 64-73).
Paltoglou, G., & Thelwall, M. (2013). Seeing stars of valence and arousal in blog posts. IEEE Transactions on Affective Computing, 4(1), 116-123.