隨著人機對話系統的快速發展，要如何讓對話系統更人性化一直是備受討論的課題，讓對話系統如同人類般擁有溝通技巧與同理心是近期的研究議題之一。
本論文之貢獻為結合同理心定義，設計具有兩種層面同理心的屬性模型與損失函數。情緒層面同理心計算使用者與系統回應兩者情緒的均方根誤差，以滿足同理心定義中受到使用者情緒感染，進而以類似、相近的情感去回應。認知層面同理心則透過交叉熵計算使用者語句與系統回應間對話內容的關聯性，滿足去設身處地了解他人觀點的同理心定義，透過高關聯性的回應讓使用者感受到其想法被他人所理解。本研究透過即插即用結構融合包含兩種同理心層面的屬性模型對語言生成模型進行同理心擾動，引導模型生成具有同理心的對話回應。
本論文使用EmpatheticDialogues作為同理對話回應生成的目標資料集，使用即插即用結構與同理心擾動應用，透過針對語言生成模型的擾動引導其生成方向。根據實驗結果，此模型在客觀評測指標中，情緒同理心層面的均方根誤差相較baseline降低0.0161，認知同理心層面的交叉熵相比baseline降低0.2424，原文相似度的BLEU分數提升0.29。在主觀人工評測方面，同理心、相關性與流暢度皆勝過baseline。因此本論文提出的基於即插即用架構進行同理心擾動應用能有效引導生成更具有同理心的對話回應句。

With the development of the dialogue system, making dialogue systems behave like humans has been discussed for a long time. How to have communication skills and empathy is one of the recent research topics in dialogue systems.
The contribution of this thesis is to combine the definition of empathy and design the attribute model and loss function of empathy in two aspects. Emotional empathy calculates the root mean square error of the emotional valence scores of the user and the system. According to the definition of empathy, the system is affected by the user's emotions, and then responds with similar emotions. Cognitive empathy uses cross-entropy to calculate the relevance of the dialogue content between the user’s sentence and the system’s response. Put yourself in others place to understand their opinions, and let users feel that their ideas are understood by others through highly relevant system responses. This study uses a plug-and-play structure combined with the empathy attribute model to perturb the dialogue generation model with empathy, and guide the model to generate an empathetic dialogue response.
This thesis uses EmpatheticDialogues as the training dataset in generating empathetic responses. Use plug-and-play structure and empathy perturbation to guide the direction of the dialogue generation. According to the objective evaluation experiment, the root mean square error of emotional empathy of our proposed method is reduced by 0.0161 compared with the baseline. The cross-entropy of the cognitive empathy is 0.2424 lower than the baseline. The BLEU score increased by 0.29. In subjective evaluation, the scores of empathy, relevance and fluency of our proposed method are better than baseline. Therefore, the application of empathy perturbation based on the plug-and-play architecture proposed in this study can effectively guide the generation of dialogue response sentences with empathy.

[1] P. Fung, A. Dey, F. B. Siddique, R. Lin, Y. Yang, Y. Wan, and H. Y. R. Chan. Zara the supergirl: An empathetic personality recognition system. in Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations. 2016.
[2] H. Rashkin, E. M. Smith, M. Li, and Y.-L. Boureau. Towards Empathetic Open-domain Conversation Models: A New Benchmark and Dataset. in Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019.
[3] R. Zandie and M. H. Mahoor. Emptransfo: A multi-head transformer architecture for creating empathetic dialog systems. in The Thirty-Third International Flairs Conference. 2020.
[4] Z. Lin, P. Xu, G. I. Winata, F. B. Siddique, Z. Liu, J. Shin, and P. Fung. Caire: An end-to-end empathetic chatbot. in Proceedings of the AAAI Conference on Artificial Intelligence. 2020.
[5] Q. Li, H. Chen, Z. Ren, Z. Chen, Z. Tu, and J. Ma, Empgan: Multi-resolution interactive empathetic dialogue generation. arXiv e-prints, 2019: p. arXiv: 1911.08698.
[6] Y.-H. Wang, 基於對話情境及同理分析於條件轉換器之同理回應生成. 成功大學資訊工程學系學位論文, 2020: p. 1-67.
[7] H. C. Yu, T. H. Huang, and H. H. Chen. Domain dependent word polarity analysis for sentiment classification. in 24th Conference on Computational Linguistics and Speech Processing, ROCLING 2012. 2012.
[8] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 2019.
[9] A. Agrawal and A. An. Unsupervised emotion detection from text using semantic and syntactic relations. in 2012 IEEE/WIC/ACM International Conferences on Web Intelligence and Intelligent Agent Technology. 2012. IEEE.
[10] S. Dathathri, A. Madotto, J. Lan, J. Hung, E. Frank, P. Molino, J. Yosinski, and R. Liu. Plug and Play Language Models: A Simple Approach to Controlled Text Generation. in International Conference on Learning Representations. 2019.
[11] J. Li, M. Galley, C. Brockett, J. Gao, and W. B. Dolan. A Diversity-Promoting Objective Function for Neural Conversation Models. in Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2016.
[12] Y. Zhang, S. Sun, M. Galley, Y.-C. Chen, C. Brockett, X. Gao, J. Gao, J. Liu, and W. B. Dolan. DIALOGPT: Large-Scale Generative Pre-training for Conversational Response Generation. in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations. 2020.
[13] A. Smith, Cognitive empathy and emotional empathy in human behavior and evolution. The Psychological Record, 2006. 56(1): p. 3-21.
[14] M. E. Peters, M. Neumann, M. Iyyer, M. Gardner, C. Clark, K. Lee, and L. Zettlemoyer. Deep contextualized word representations. in Proceedings of NAACL-HLT. 2018.
[15] K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu. Bleu: a method for automatic evaluation of machine translation. in Proceedings of the 40th annual meeting of the Association for Computational Linguistics. 2002.
[16] T. Wen, M. Gašić, N. Mrkšić, P. Su, D. Vandyke, and S. Young. Semantically conditioned lstm-based Natural language generation for spoken dialogue systems. in Conference Proceedings-EMNLP 2015: Conference on Empirical Methods in Natural Language Processing. 2015.
[17] C.-W. Liu, R. Lowe, I. Serban, M. Noseworthy, L. Charlin, and J. Pineau. How NOT To Evaluate Your Dialogue System: An Empirical Study of Unsupervised Evaluation Metrics for Dialogue Response Generation. in EMNLP. 2016.
[18] R. Socher, A. Perelygin, J. Wu, J. Chuang, C. D. Manning, A. Y. Ng, and C. Potts. Recursive deep models for semantic compositionality over a sentiment treebank. in Proceedings of the 2013 conference on empirical methods in natural language processing. 2013.
[19] Z. Lin, A. Madotto, J. Shin, P. Xu, and P. N. Fung. MoEL: Mixture of Empathetic Listeners. in EMNLP-IJCNLP 2019-2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint Conference on Natural Language Processing, Proceedings of the Conference. 2019.
[20] J. Shin, P. Xu, A. Madotto, and P. Fung, Happybot: Generating empathetic dialogue responses by improving user experience look-ahead. arXiv preprint arXiv:1906.08487, 2019.
[21] S. Mohammad. Obtaining reliable human ratings of valence, arousal, and dominance for 20,000 English words. in Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2018.