隨著線上社群的快速發展，已然成為了現代人生活不可或缺的一環。藉由在社群中分享訊息與回應，讓使用者獲得不同程度的情緒抒發。然而，由於社群成員相當複雜，參與動機也各有不同，使得社群管理面臨嚴峻的挑戰。為維持社群的長遠發展，首要之務便是了解社群成員的情緒變化，以建構正面的社群氛圍。然而，成員的情緒變化瞬息萬變，顯然無法以人工審閱方式來達成。因此，使用機器學習方法自動偵測社群發文情緒，便成為可行的解決之道。有鑑於此，本論文將以多維度情緒沙漏模型為基礎，配合深度學習方法發展有效之多層次多模型之情緒偵測架構。為此，我們提出了一套以文章極性偵測為前處理之雙層的偵測架構: 第一層用於判別文章情緒之正負極性；第二層則用於產生解析度更高之情緒值。在第一層中，我們建構了五個不同特質的二分類模型，分別以連續過濾(Filtered-NP)與平衡過濾(Balanced-NP)方式加以組合運用，以提升文章極性辨識準確性。在第二層中，我們提出了NP-MaxCount與NP-Avg二種架構，分別以加權平均與投票方式組合多個不同的二分類模型的輸出。為驗證提出方法之有效性，我們使用IMDB影評資料集進行實驗。結果顯示，當第一層採用Filtered-NP方法時，可產生比傳統單一極性偵測模型更準確之分類結果。此外，當第二層使用NP-Avg與NP-MaxCount架構時，結果確實優於傳統直覺式作法。當使用Filtered-NP-Avg組合架構時，便可獲得最佳之偵測準確性。上述實驗結果驗證了本研究提出方法之有效性，並可做為網站管理者重要參考依據。

With the rapid development of online communities, they have become an indispensable part of modern life. By sharing information and responding within these communities, users experience varying degrees of emotional expression. However, the complexity of community members and their diverse motivations present significant challenges for community management. To ensure sustainable community development, it is crucial to understand members' emotional dynamics to foster a positive community atmosphere. Given the rapid fluctuations in emotions, manual review is clearly insufficient. As such, employing machine learning methods to automatically detect emotional expressions in community posts becomes a viable solution.

    In light of this, this study develops an effective multi-level and multi-model emotional detection framework based on the multi-dimensional Hourglass Model of Emotions, combined with deep learning techniques. Specifically, we propose a two-layer detection architecture with sentiment polarity detection as a preprocessing step. The first layer distinguishes the positive or negative polarity of a post, while the second layer generates more granular emotional values.

    In the first layer, we construct five binary classification models with distinct characteristics, employing two combination strategies: Filtered-NP (continuous filtering) and Balanced-NP
 (balanced filtering), to enhance sentiment polarity classification accuracy. In the second layer, we introduce two architectures, NP-MaxCount and NP-Avg, which respectively use weighted averaging and voting mechanisms to integrate the outputs of multiple binary classification models.

    To validate the effectiveness of the proposed methods, experiments were conducted using the IMDB movie review dataset. The results show that when the Filtered-NP method is applied in the first layer, it yields more accurate classification results compared to traditional single-polarity detection models. Moreover, when NP-Avg or NP-MaxCount is used in the second layer, the outcomes outperform conventional heuristic approaches. The best detection accuracy is achieved by combining Filtered-NP and NP-Avg architectures.

    These experimental results demonstrate the validity of the proposed methods and provide valuable insights for website administrators.

目錄
第一章 緒論	1
第二章 文獻探討與相關技術	4
2.1 情緒模型(Emotions Models)	4
2.2 深度學習技術	6
2.3 多模型多階段之分類架構	7
第三章 以多模型為基礎之多維度情緒預測方法	10
3.1 問題陳述	10
3.2 以正負極性判別為前處理之多模型情緒值預測架構	11
3.2.1  Naïve-C6 Framework	11
3.2.2  Naïve-NP-C3 Framework	13
3.2.3  Naïve-NP-MaxCount Framework	13
3.2.4  Naïve-NP-Avg Framework	15
3.3	Filtered-NP Framework	16
3.4	Balanced-NP Framework	18
第四章 實驗結果與討論	21
4.1 實驗設定	21
4.1.1 多維度情緒資料集之建立	21
4.1.2 塑模時之參數設定	23
4.2 情緒預測架構之效能驗證	24
4.2.1 多模型之建立流程	25
4.2.2 實驗結果比較	26
4.3 多模型架構之細部效能比較	28
4.3.1 極性判別方法之比較	28
4.3.2 正負極性資料之準確性比較	28
4.3.3 多模型架構成本效益之驗證	30
4.4 研究限制	30
第五章 結論與未來工作	32
參考文獻	33

圖目錄
圖2-1  Arousal-Valence情緒模型 	4
圖2-2  Revised Hourglass of Emotions 	6
圖2-3  BERT之pre-training與Fine-Tuning概念圖 	6
圖2-4  Transformer之模型架構	7
圖2-5  以連續過濾法進行線上拍賣之詐騙偵測	8
圖2-6  使用多模型架構進行線上拍賣之詐騙偵測	8
圖3-1  將文章進行多維度多程度值之情緒偵測	11
圖3-2  Naïve-C6 Framework 	12
圖3-3  單一情緒維度使用單一模型進行6分類之預測結果	12
圖3-4  單一BERT模型對Introspection維度之正負極性預測結果	13
圖3-5  Naïve BERT-NP-C3多模型文章情緒預測架構(單維度)	13
圖3-6  BERT-NP-MaxCount多模型文章情緒預測架構	13
圖3-7  BERT-NP Avg情緒預測架構(單維度)	16
圖3-8  以過濾法為基礎之正負情緒處理架構	17
圖3-9  Filter-NP情緒預測架構	17
圖3-10  Filter-based BERT-NP-MaxCount預測架構	18
圖3-11  Filter-based BERT-NP-Avg預測架構	18
圖3-12  Balanced BERT-NP-C3情緒偵測架構	19
圖3-13  Balanced BERT-NP-MaxCount(左)與Balanced-BERT-NP-Avg(右)情緒偵測架構	20
圖4-1 為使生成式AI系統自動標示文章情緒所提供之提示前文	22
圖4-2 由Gpt-4-turbo-preview 協助產生之IMDB多維度情緒資料集	23

表目錄
表2-1  沙漏模型之各維度定義與代表性情緒詞	5
表3-1  情緒沙漏模型之多維度情緒值對應表	10
表3-2  BERT-NP-MaxCount預測架構之運作流程虛擬碼 	14
表3-3  以不同資料配比建構之極性判別模型效能比較 	17
表3-4  BERT-NP(1:6)與BERT-NP(6:1)之模型效能 	17
表4-1  生成式AI標註情緒標籤與人工標籤之差異比較 	22
表4-2  本研究使用之軟體套件	23
表4-3  訓練集資料配比	23
表4-4  驗證集資料配比 	24
表4-5  測試集資料配比 	24
表4-6  訓練、驗證與測試集之資料筆數 	24
表4-7  以正負極性為前處理之各種預測架構效能比較	26
表4-8  以Filtered-NP正負極性判別為基礎之預測架構效能比較	27
表4-9  以Balanced -NP正負極性判別為基礎之預測架構效能比較	27
表4-10  以不同極性前處理方法獲得之情緒值預測結果比較	27
表4-11  Naïve- NP、Filtered- NP與Balanced- NP極性判別方法之準確率比較	28
表4-12  以各種預測架構之正負極性資料預測結果比較	29
表4-13  論文中各單一模型之塑模時間統計	30
表4-14  以IMDB資料集建立之多模型架構預測Yelp資料集之實驗結果	31

1.	Acheampong, F. A., Chen W., Henry N.-M. (2020), "Text-based emotion detection: Advances, challenges, and opportunities", Engineering Reports, Volume 2, Issue 7, Jul 2020.
2.	Bandhakavi, A., et al., (2017) "Lexicon based feature extraction for emotion text classification," Pattern Recognition Letters, vol. 93, 2017, pp. 133-142.
3.	Chen, B., (2021) "Sentiment Analysis from Machine Learning to Deep Learning,” 2021 International Conference on Electronic Information Engineering and Computer Science, pp. 724-728.
4.	Chen, C.-H., Lee, W.-P., Huang, J.-Y., (2018) "tracking and recognizing emotions in short text messages from online chatting services," Information Processing and management, vol. 54, 2018, pp. 1325-1344.
5.	Devlin, Jacob, Ming-Wei Chang, Kenton Lee, Kristina Toutanova, (2019) “BERT: Pre- training of Deep Bidirectional Transformers for Language Understanding,” last revised 24 May 2019 (v2), arXiv:1810.04805.
6.	Dorottya Demszky, Dana Movshovitz-Attias, Jeongwoo Ko, Alan Cowen, Gaurav Nemade, Sujith Ravi,(2020) “ GoEmotions: A Dataset of Fine-Grained Emotions, ” Submitted on 1 May 2020 (v1), last revised 3 Jun 2020 (this version, v2), arXiv:2005.00547.
7.	Ekman, P., & Friesen, W. V. (1976). Measuring facial movement. Environmental Psychology & Nonverbal Behavior, 1(1), 56–75.
8.	Huang, Z. and Fang, Z., (2020) "An Entity-Level Sentiment Analysis of Financial Text Based on Pre-Trained Language Model," 2020 IEEE 18th International Conference on Industrial Informatics (INDIN), pp. 391-396.
9.	Hu, P., et al, (2017) "Valence-Arousal Analysis for Mental-Health Document Retrieval," 2017 International Conference on Orange Technologies (ICOT), pp. 61-64.
10.	Khatua, A., et al. (2019), “Tweeting in Support of LGBT?: A Deep Learning Approach,” Proceedings of the ACM India Joint International Conference on Data Science and Management of Data, pp. 342–345.
11.	Kumar, A. J., Tina Esther Trueman, Erik Cambria (2021), “A Convolutional Stacked Bidirectional LSTM with a Multiplicative Attention Mechanism for Aspect Category and Sentiment Detection,” Cognitive Computation volume 13, pages1423–1432.
12.	McCormick, Chris, (2019) "BERT Research - Ep. 1 - Key Concepts & Sources, " 2019/11/1, URL:https://mccormickml.com/2019/11/11/BERT-research-ep-1-key-concepts-and-sources/
13.	Nathanson, Donald L. (1992), "Shame and Pride: Affect, Sex, and the Birth of the Self", New York: W.W. Norton, ISBN 0-393-03097-0.
14.	OOSGA(2023), “2023年TikTok用戶分佈與成長前景如何？主要趨勢為何？” https://zh.oosga.com/tiktok-overview/, Last Retrieved 2024/10/29.
15.	Plutchik, R. (2001), "The Nature of Emotions," American Scientist Vol. 89, No. 4, pp. 344- 350.
16.	Plutchik (2001), The Nature of Emotions, American Scientist, 89(4), pp. 344 - 350, 2001.
17.	Poria, S., et al. (2015), "Sentiment Data Flow Analysis by Means of Dynamic Linguistic Patterns," IEEE Computational Intelligence Magazine, Volume: 10, Issue: 4, pp. 26-36.
18.	Plutchik, R. (2001), "The Nature of Emotions," American Scientist Vol. 89, No. 4, pp. 344- 350.
19.	Kumar, S., Khuntia, J. R., & Khatua, K. K. (2021). Prediction of flow resistance in an open channel over movable beds using artificial neural network. Journal of Hydrologic Engineering, 26(5), 04021015.
20.	Socher, R., Perelygin, A., Wu, J., Chuang, J., Manning, C. D., Ng, A. Y., & Potts, C. (2013, October). Recursive deep models for semantic compositionality over a sentiment treebank. In Proceedings of the 2013 conference on empirical methods in natural language processing (pp. 1631-1642).
21.	Susanto, Y., Livingstone, A. G., Ng, B. C., & Cambria, E. (2020). The hourglass model revisited. IEEE Intelligent Systems, 35(5), 96-102.
22.	Poria, S., Cambria, E., & Gelbukh, A. (2015, September). Deep convolutional neural network textual features and multiple kernel learning for utterance-level multimodal sentiment analysis. In Proceedings of the 2015 conference on empirical methods in natural language processing (pp. 2539-2544).
23.	Purba, S. A., Tasnim, S., Jabin, M., Hossen, T., & Hasan, M. K. (2021, February). Document level emotion detection from bangla text using machine learning techniques. In 2021 International Conference on Information and Communication Technology for Sustainable Development (ICICT4SD) (pp. 406-411). IEEE.
24.	Huang, Z., & Fang, Z. (2020, July). An Entity-Level Sentiment Analysis of Financial Text Based on Pre-Trained Language Model. In 2020 IEEE 18th International Conference on Industrial Informatics (INDIN) (Vol. 1, pp. 391-396). IEEE.
25.	Acheampong, R. A., Siiba, A., Okyere, D. K., & Tuffour, J. P. (2020). Mobility-on-demand: An empirical study of internet-based ride-hailing adoption factors, travel characteristics and mode substitution effects. Transportation Research Part C: Emerging Technologies, 115, 102638.
26.	Plutchik, R. (2001). The nature of emotions: Human emotions have deep evolutionary roots, a fact that may explain their complexity and provide tools for clinical practice. American scientist, 89(4), 344-350.
27.	Tomkins, S. (1962). Affect imagery consciousness: Volume I: The positive affects. Springer publishing company.
28.	Whissell, C. M. (1989). The dictionary of affect in language. In The measurement of emotions (pp. 113-131). Academic Press.
29.	Cambria, E., Livingstone, A., & Hussain, A. (2012). The hourglass of emotions. In Cognitive behavioural systems: COST 2102 international training school, dresden, Germany, February 21-26, 2011, revised selected papers (pp. 144-157). Springer Berlin Heidelberg.
30.	Vaswani, A. (2017). Attention is all you need. Advances in Neural Information Processing Systems.
31.	張昭憲, & 陳世軒. (2021). 以模型融合為基礎之線上拍賣詐騙偵測. 資訊管理學報, 28(4), 419-444.
32.	田家豪. (2022). 線上社群發文之複合情緒分析. 淡江大學資訊管理學系碩士班學位論文, 2022, 1-36.
33.	李世傑. (2024). 以深度學習進行文章層級之多維度情緒分析. 淡江大學資訊管理學系碩士班學位論文, 2024, 1-43.