自動化的對話系統被運用於許多領域，現階段最熱門的應用便是開放域對話，而要如何讓使用者提高對對話系統的滿意度始終都是最重要的課題，也為此作了許多研究，其中的一個分支是同理心對話系統，而影響同理心最重要的因素是情緒，因此幾乎所有研究都會圍繞情緒進行。然而這些研究所考慮的情緒都是一則對話的整體情緒，而非對話內每一句話語的情緒。一般來說，即使有一個情緒可以概括整個對話，但這並不代表對話內部所有的話語都會是同樣的情緒，而且根據心理學的研究，能有效地跟隨一個對話的情緒流將會有助於同理心的表現。
因此本研究認為若是能將話語情緒轉移的關係加入進對話系統，那麼就可以提高對話系統的同理心表達能力。為此本研究提出了「multi-task」、「table」、「matrix」等三種方法來實作情緒轉移的效果，並且設計了兩個主要實驗，分別用來驗證話語間情緒轉移的有效性和泛用性。第一個實驗會在選定的兩個經典同理心對話架構CEM和RecEC之中，透過提出的方法將情緒轉移的要素加進模型之中來驗證有效性。第二個實驗會是用不同的資料集來驗證情緒轉移的泛用性。最後根據實驗結果，情緒轉移的要素不管在有效性還是泛用性上都得到了證實，在同理心的自動指標上都相較於原始模型擁有更好的結果，顯示情緒轉移對於同理心對話確實有正向的影響。

Automated dialogue systems are being applied in various fields. The most popular application is open-domain conversation, and how to enhance user satisfaction with these systems is always a pivotal challenge. As a result, numerous studies have been conducted, with one branch focusing on empathetic dialogue systems. Emotion is the most important factor influencing empathy, and therefore, nearly all research revolves around emotions. However, the emotions considered in these studies typically pertain to the overall dialogue emotion rather than the emotion of each individual utterance. Generally, even if one emotion can summarize the entire conversation, it doesn't imply that all utterances within the dialogue share the same emotion.
Therefore, this study proposes that by incorporating the concept of emotion transition between utterances into dialogue systems, the empathetic expression capability of the systems can be enhanced. To achieve this, three methods are introduced: "multi-task," "table," and "matrix," aiming to implement emotion transition. Two experiments are designed to validate the effectiveness and generalizability of emotion transition. In the first experiment, the proposed methods are integrated into two empathetic dialogue frameworks, CEM and RecEC, to verifies effectiveness. The second experiment aims to verifies the generalizability of emotion transition using different datasets. Based on experimental results, the factor of emotion transition has been validated in terms of both effectiveness and generalizability. The models incorporating emotion transition outperform the original models in terms of automatic empathy metrics, demonstrating a positive effect of emotion transition on empathetic dialogues system.

論文審定書 i
摘要 ii
Abstract iii
目錄 iv
圖次 vii
表次 viii
第一章 緒論 1
1.1 研究背景和動機 1
1.2 研究目的 2
第二章 文獻探討 3
2.1 對話系統(Dialogue System) 3
2.1.1 問答系統(Question Answer System) 3
2.1.2 任務導向對話系統(Task-oriented Dialogue System) 4
2.2 開放域對話系統(Open-Domain Dialogue System) 4
2.2.1 基於檢索的對話系統(Retrieval-based Dialogue System) 5
2.2.2 基於生成的對話系統(Generation-based Dialogue System) 6
2.3 情感對話系統(Emotional Dialogue System) 7
2.4 同理心 10
2.5 同理心對話系統(Empathetic Dialogue System) 11
2.5.1 Commonsense-aware Empathetic Chatting Machine(CEM)[24] 11
2.5.2 RecEC[25] 13
第三章 研究方法 16
3.1 研究流程與實驗設計 16
3.2 基底模型 18
3.3 情緒轉移 18
3.3.1 Multi-Task 18
3.3.2 Table 20
3.3.3 Matrix 22
3.4 資訊結合方法 24
第四章 實驗結果與分析 26
4.1 資料集介紹 26
4.1.1 Empathetic-Dialogues[27] 26
4.1.2 ESConv[28] 28
4.1.3 資料前處理 29
4.2 評估方法 30
4.3 實驗流程 32
4.4 情緒轉移對於同理心對話系統之影響 32
4.4.1與原始模型之間的差異 32
4.4.2 情緒轉移方法之間的差異 34
4.5 情緒轉移在不同資料集所能發揮的效用 35
4.6 其他差異實驗的綜合討論 37
4.6.1 生成時融合策略差異 37
4.6.2 對話情緒與話語情緒的差異 38
4.6.3 考慮複數句話語情緒的差異 39
4.7 對話生成範例 39
第五章 結論 43
5.1 結論 43
5.2 未來展望 43
參考文獻 45
附錄一 51

[1].Sun, B., & Li, K. (2021). Neural dialogue generation methods in open domain: a survey. Natural Language Processing Research, 1(3-4), 56-70.
[2].Zhao, Y. J., Li, Y. L., & Lin, M. (2019, July). A review of the research on dialogue management of task-oriented systems. In Journal of Physics: Conference Series (Vol. 1267, No. 1, p. 012025). IOP Publishing.
[3].Raamkumar, A. S., & Yang, Y. (2022). Empathetic Conversational Systems: A Review of Current Advances, Gaps, and Opportunities. IEEE Transactions on Affective Computing.
[4].Chen, H., Liu, X., Yin, D., & Tang, J. (2017). A survey on dialogue systems: Recent advances and new frontiers. Acm Sigkdd Explorations Newsletter, 19(2), 25-35.
[5].Guthridge, M., & Giummarra, M. J. (2021). The taxonomy of empathy: A meta-definition and the nine dimensions of the empathic system. Journal of Humanistic Psychology, 00221678211018015.
[6].Nabi, R. L. (2015). Emotional flow in persuasive health messages. Health communication, 30(2), 114-124.
[7].Weilenmann, S., Schnyder, U., Parkinson, B., Corda, C., Von Kaenel, R., & Pfaltz, M. C. (2018). Emotion transfer, emotion regulation, and empathy-related processes in physician-patient interactions and their association with physician well-being: a theoretical model. Frontiers in psychiatry, 9, 389.
[8].Allam, A. M. N., & Haggag, M. H. (2012). The question answering systems: A survey. International Journal of Research and Reviews in Information Sciences (IJRRIS), 2(3).
[9].Huang, M., Zhu, X., & Gao, J. (2020). Challenges in building intelligent open-domain dialog systems. ACM Transactions on Information Systems (TOIS), 38(3), 1-32.
[10].Zhao, T., Zhao, R., & Eskenazi, M. (2017). Learning discourse-level diversity for neural dialog models using conditional variational autoencoders. arXiv preprint arXiv:1703.10960.
[11].Serban, I., Sordoni, A., Lowe, R., Charlin, L., Pineau, J., Courville, A., & Bengio, Y. (2017, February). A hierarchical latent variable encoder-decoder model for generating dialogues. In Proceedings of the AAAI conference on artificial intelligence (Vol. 31, No. 1).
[12].Qian, Q., Huang, M., Zhao, H., Xu, J., & Zhu, X. (2018, July). Assigning Personality/Profile to a Chatting Machine for Coherent Conversation Generation. In Ijcai (pp. 4279-4285).
[13].Ma, Z., Dou, Z., Zhu, Y., Zhong, H., & Wen, J. R. (2021, July). One chatbot per person: Creating personalized chatbots based on implicit user profiles. In Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval (pp. 555-564).
[14].Li, J., Monroe, W., Ritter, A., Galley, M., Gao, J., & Jurafsky, D. (2016). Deep reinforcement learning for dialogue generation. arXiv preprint arXiv:1606.01541.
[15].Ochs, M., Pelachaud, C., & Sadek, D. (2008, May). An empathic virtual dialog agent to improve human-machine interaction. In Proceedings of the 7th international joint conference on Autonomous agents and multiagent systems-Volume 1 (pp. 89-96).
[16].Zhang, Y., Galley, M., Gao, J., Gan, Z., Li, X., Brockett, C., & Dolan, B. (2018). Generating informative and diverse conversational responses via adversarial information maximization. Advances in Neural Information Processing Systems, 31.
[17].Majumder, N., Hong, P., Peng, S., Lu, J., Ghosal, D., Gelbukh, A., ... & Poria, S. (2020). MIME: MIMicking emotions for empathetic response generation. arXiv preprint arXiv:2010.01454.
[18].Song, Z., Zheng, X., Liu, L., Xu, M., & Huang, X. J. (2019, July). Generating responses with a specific emotion in dialog. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (pp. 3685-3695).
[19].Zhou, H., Huang, M., Zhang, T., Zhu, X., & Liu, B. (2018, April). Emotional chatting machine: Emotional conversation generation with internal and external memory. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 32, No. 1).
[20].Chen, M. Y., Li, S., & Yang, Y. (2022). EmpHi: Generating Empathetic Responses with Human-like Intents. arXiv preprint arXiv:2204.12191.
[21].Naous, T., Antoun, W., Mahmoud, R. A., & Hajj, H. (2021). Empathetic BERT2BERT conversational model: Learning Arabic language generation with little data. arXiv preprint arXiv:2103.04353.
[22].Wang, Y. H., Hsu, J. H., Wu, C. H., & Yang, T. H. (2021, January). Transformer-based empathetic response generation using dialogue situation and advanced-level definition of empathy. In 2021 12th International Symposium on Chinese Spoken Language Processing (ISCSLP) (pp. 1-5). IEEE.
[23].Li, Q., Li, P., Ren, Z., Ren, P., & Chen, Z. (2022, June). Knowledge bridging for empathetic dialogue generation. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 36, No. 10, pp. 10993-11001).
[24].Sabour, S., Zheng, C., & Huang, M. (2022, June). Cem: Commonsense-aware empathetic response generation. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 36, No. 10, pp. 11229-11237).
[25].Gao, J., Liu, Y., Deng, H., Wang, W., Cao, Y., Du, J., & Xu, R. (2021, November). Improving empathetic response generation by recognizing emotion cause in conversations. In Findings of the association for computational linguistics: EMNLP 2021 (pp. 807-819).
[26].Shen, L., Zhang, J., Ou, J., Zhao, X., & Zhou, J. (2021). Constructing emotion consensus and utilizing unpaired data for empathetic dialogue generation. arXiv preprint arXiv:2109.07779.
[27].Rashkin, H., Smith, E. M., Li, M., & Boureau, Y. L. (2018). Towards empathetic open-domain conversation models: A new benchmark and dataset. arXiv preprint arXiv:1811.00207.
[28].Liu, S., Zheng, C., Demasi, O., Sabour, S., Li, Y., Yu, Z., ... & Huang, M. (2021). Towards emotional support dialog systems. arXiv preprint arXiv:2106.01144.
[29].Welivita, A., Xie, Y., & Pu, P. (2021, November). A large-scale dataset for empathetic response generation. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (pp. 1251-1264).
[30].Liu, Y., Maier, W., Minker, W., & Ultes, S. (2022, November). Empathetic dialogue generation with pre-trained RoBERTa-GPT2 and external knowledge. In Conversational AI for Natural Human-Centric Interaction: 12th International Workshop on Spoken Dialogue System Technology, IWSDS 2021, Singapore (pp. 67-81). Singapore: Springer Nature Singapore.
[31].Ruder, S. (2017). An overview of multi-task learning in deep neural networks. arXiv preprint arXiv:1706.05098.
[32].Zheng, C., Liu, Y., Chen, W., Leng, Y., & Huang, M. (2021). Comae: A multi-factor hierarchical framework for empathetic response generation. arXiv preprint arXiv:2105.08316.
[33].Ching, W. K., & Ng, M. K. (2006). Markov chains. Models, algorithms and applications.
[34].Goleman, D. (2008). Hot to help: When can empathy move us to action. Greater Good Magazine, 1.
[35].Demszky, D., Movshovitz-Attias, D., Ko, J., Cowen, A., Nemade, G., & Ravi, S. (2020). GoEmotions: A dataset of fine-grained emotions. arXiv preprint arXiv:2005.00547.
[36].Bosselut, A., Rashkin, H., Sap, M., Malaviya, C., Celikyilmaz, A., & Choi, Y. (2019). COMET: Commonsense transformers for automatic knowledge graph construction. arXiv preprint arXiv:1906.05317.
[37].Wikipedia contributors. (2023, August 13). Perplexity. In Wikipedia, The Free Encyclopedia. Retrieved 03:53, August 21, 2023, from https://en.wikipedia.org/w/index.php?title=Perplexity&oldid=1170145864
[38].Li, J., Galley, M., Brockett, C., Gao, J., & Dolan, B. (2015). A diversity-promoting objective function for neural conversation models. arXiv preprint arXiv:1510.03055.
[39].Papineni, K., Roukos, S., Ward, T., & Zhu, W. J. (2002, July). Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics (pp. 311-318).
[40].Zhang, T., Kishore, V., Wu, F., Weinberger, K. Q., & Artzi, Y. (2019). Bertscore: Evaluating text generation with bert. arXiv preprint arXiv:1904.09675.
[41].Keshtkar, F., & Inkpen, D. (2011, October). A pattern-based model for generating text to express emotion. In International Conference on Affective Computing and Intelligent Interaction (pp. 11-21). Berlin, Heidelberg: Springer Berlin Heidelberg.
[42].Asghar, N., Poupart, P., Hoey, J., Jiang, X., & Mou, L. (2018). Affective neural response generation. In Advances in Information Retrieval: 40th European Conference on IR Research, ECIR 2018, Grenoble, France, March 26-29, 2018, Proceedings 40 (pp. 154-166). Springer International Publishing.
[43].Ghosh, S., Chollet, M., Laksana, E., Morency, L. P., & Scherer, S. (2017). Affect-lm: A neural language model for customizable affective text generation. arXiv preprint arXiv:1704.06851.
[44]. Sutskever, I., Vinyals, O., & Le, Q. V. (2014). Sequence to sequence learning with neural networks. Advances in neural information processing systems, 27.