结合微调与重排序的情感可控对话生成方法

doi:10.3778/j.issn.1673-9418.2106060

摘要/Abstract

摘要： 现有的情感对话生成方法通常以基于序列到序列（Seq2Seq）的对话生成模型为基础，在编码或解码进行情感方面的改进，该类方法虽然能够实现一定的情感回复生成能力，但容易出现低质量的通用回复问题。为解决以上问题，实现情感可控且高质量的回复生成，提出了一种结合微调与重排序的情感可控对话生成方法，称为情感生成式预训练Transformer（EmoGPT）。在模型训练阶段，提出使用带有情感类别标签的对话语料微调GPT-2模型的方法，使其学习语句中语义和情感的依赖关系；在回复生成阶段，提出情感重排序策略对生成的多句回复进行情感打分并排序，以提高回复情感可控性。在使用带有情感标签的对话数据集的情感回复生成实验结果显示，带有情感重排序策略的EmoGPT在生成回复的内容相关性和情感一致性方面取得了领先于对比模型的性能，从而验证了文中方法的情感可控且高质量的回复生成能力。

关键词: 对话回复生成, 情感可控, 微调, 情感重排序, 内容相关性, 情感一致性

Abstract: The existing emotional dialogue generation methods are usually based on the sequence-to-sequence (Seq2Seq) dialogue generation models, and improve the encoding or decoding in terms of emotion. This method can achieve a certain extent of emotional response generation abilities but is prone to low-quality and common response problems. To solve these problems and achieve emotion controllable and high-quality response generation, this paper proposes an emotion controllable dialogue generation method combining fine-tuning and reranking, called EmoGPT (emotional generative pre-trained transformer). In the model training phase, a method for fine-tuning the GPT-2 model using dialogue corpus with emotion category labels is proposed to learn the semantic and emotional dependence of the sentences; in the response generation phase, an emotion reranking strategy is proposed to score and sort the generated multiple responses to improve the controllability of the response emotions. Experimental results of emotional response generation using a dialogue dataset with emotion labels show that EmoGPT with an emotion reranking strategy achieves better performance than comparison models in terms of content relevance and emotion consistency of the generated responses, thus verifying the ability of the proposed method to generate emotion controllable and high-quality responses.

Key words: dialogue response generation, emotion controllable, fine-tuning, emotion reranking, content relevance, emotion consistency

杜宝祥, 马志强, 王春喻, 贾文超, 王洪彬. 结合微调与重排序的情感可控对话生成方法[J]. 计算机科学与探索, 2023, 17(4): 953-963.

DU Baoxiang, MA Zhiqiang, WANG Chunyu, JIA Wenchao, WANG Hongbin. Emotion Controllable Dialogue Generation Method Combining Fine-Tuning and Reranking[J]. Journal of Frontiers of Computer Science and Technology, 2023, 17(4): 953-963.

参考文献

[1] SUTSKEVER I, VINYALS O, LE Q V. Sequence to sequ-ence learning with neural networks[J]. arXiv:1409.3215, 2014.
[2] ZHOU H, HUANG M, ZHANG T, et al. Emotional chat-ting machine: emotional conversation generation with inter-nal and external memory[C]//Proceedings of the 32nd AAAI Conference on Artificial Intelligence, the 30th Innovative Applications of Artificial Intelligence, and the 8th AAAI Symposium on Educational Advances in Artificial Intelli-gence, New Orleans, Feb 2-7, 2018. Menlo Park: AAAI, 2018: 730-738.
[3] SHANTALA R, KYSELOV G, KYSELOVA A. Neural dia-logue system with emotion embeddings[C]//Proceedings of the 2018 IEEE 1st International Conference on System Anal-ysis & Intelligent Computing, Kyiv, Oct 8-12, 2018. Pisca-taway: IEEE, 2018: 1-4.
[4] HUANG C, ZAIANE O R, TRABELSI A, et al. Automatic dialogue generation with expressed emotions[C]//Procee-dings of the 2018 Conference of the North American Cha-pter of the Association for Computational Linguistics: Human Language Technologies, New Orleans, Jun 1-6, 2018. Strou-dsburg: ACL, 2018: 49-54.
[5] ZHANG R, WANG Z, MAI D. Building emotional convers-ation systems using multi-task Seq2Seq learning[C]//LNCS 10619: Proceedings of the 6th CCF Conference on Natural Language Processing and Chinese Computing, Dalian, Nov 8-12, 2017. Cham: Springer, 2017: 612-621.
[6] SONG Z, ZHENG X, LIU L, et al. Generating responses with a specific emotion in dialog[C]//Proceedings of the 57th Conference of the Association for Computational Lin-guistics, Florence, Jul 28-Aug 2, 2019. Stroudsburg: ACL, 2019: 3685-3695.
[7] XU W, GU X, CHEN G. Generating emotional controllable response based on multi-task and dual attention framework[J]. IEEE Access, 2019, 7: 93734-93741.
[8] KONG X, LI B, NEUBIG G, et al. An adversarial approach to high-quality, sentiment-controlled neural dialogue gene-ration[J]. arXiv:1901.07129, 2019.
[9] PENG D, ZHOU M, LIU C, et al. Human-machine dialogue modelling with the fusion of word-and sentence-level emot-ions[J]. Knowledge-Based Systems, 2020, 192: 105319.
[10] MA Z, YANG R, DU B, et al. A control unit for emotional conversation generation[J]. IEEE Access, 2020, 8: 43168-43176.
[11] SHEN L, FENG Y. CDL: curriculum dual learning for emotion-controllable response generation[C]//Proceedings of the 58th Annual Meeting of the Association for Compu-tational Linguistics, Jul 5-10, 2020. Stroudsburg: ACL, 2020: 556-566.
[12] RADFORD A, WU J, CHILD R, et al. Language models are unsupervised multitask learners[J]. OpenAI Blog, 2019, 1(8): 9.
[13] VINYALS O, LE Q. A neural conversational model[J]. arXiv:1506.05869, 2015.
[14] CHO K, VAN MERRI?NBOER B, GULCEHRE C, et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation[J]. arXiv:1406.1078, 2014.
[15] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Proceedings of the Annual Conference on Neural Information Processing Systems 2017, Long Beach, Dec 4-9, 2017. Red Hook: Curran Associates, 2017: 5998-6008.
[16] ZHANG Y, SUN S, GALLEY M, et al. DIALOGPT: large-scale generative pretraining for conversational response gene-ration[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demon-strations, Jul 5-10, 2020. Stroudsburg: ACL, 2020: 270-278.
[17] LI X, LI P, BI W, et al. Relevance-promoting language model for short-text conversation[C]//Proceedings of the 34th AAAI Conference on Artificial Intelligence, the 32nd Innovative Applications of Artificial Intelligence Confer-ence, the 10th AAAI Symposium on Educational Advances in Artificial Intelligence, New York, Feb 7-12, 2020. Menlo Park: AAAI, 2020: 8253-8260.
[18] OLUWATOBI O, MUELLER E. DLGNet: a transformer-based model for dialogue response generation[J]. arXiv:1908.01841, 2019.
[19] HOLTZMAN A, BUYS J, DU L, et al. The curious case of neural text degeneration[J]. arXiv:1904.09751, 2019.
[20] LI J T, YAN R. Overview of the NLPCC 2018 shared task: multi-turn human-computer conversations[C]//LNCS 11109: Proceedings of the 7th CCF International Conference on Natural Language Processing and Chinese Computing, Hohhot, Aug 26-30, 2018. Cham: Springer, 2018: 446-451.
[21] DEVLIN J, CHANG M W, LEE K, et al. BERT: pre-training of deep bidirectional transformers for language understanding[C]//Proceedings of the 2019 Conference of the North American Chapter of the Association for Compu-tational Linguistics: Human Language Technologies, Min-neapolis, Jun 2-7, 2019. Stroudsburg: ACL, 2019: 4171-4186.
[22] SERBAN I V, SORDONI A, BENGIO Y, et al. Building end-to-end dialogue systems using generative hierarchical neural network models[C]//Proceedings of the 30th AAAI Conference on Artificial Intelligence, Phoenix, Feb 12-17, 2016. Menlo Park: AAAI, 2016: 3776-3784.
[23] SERBAN I V, SORDONI A, LOWE R, et al. A hierarchical latent variable encoder-decoder model for generating dialo-gues[C]//Proceedings of the 31st AAAI Conference on Artificial Intelligence, San Francisco, Feb 4-9, 2017. Menlo Park: AAAI, 2017: 3295-3301.
[24] LIN C Y. ROUGE: a package for automatic evaluation of summaries[C]//Proceedings of the 2004 Workshop on Text Summarization of ACL, Barcelona, Jul 21-26, 2004. Stroud-sburg: ACL, 2004: 74-81.
[25] WIETING J, BANSAL M, GIMPEL K, et al. Towards universal paraphrastic sentence embeddings[J]. arXiv:1511.08198, 2015.
[26] RUS V, LINTEAN M. An optimal assessment of natural language student input using word-to-word similarity met-rics[C]//LNCS 7315: Proceedings of the 11th International Conference on Intelligent Tutoring Systems, Crete, Jun 14-18, 2012. Berlin, Heidelberg: Springer, 2012: 675-676.
[27] FORGUES G, PINEAU J, LARCHEVêQUE J M, et al. Bootstrapping dialog systems with word embeddings[C]// Proceedings of the NIPS, Modern Machine Learning and Natural Language Processing Workshop, Montreal, Dec 8-13, 2014. Red Hook: Curran Associates, 2014: 2.
[28] RONG X. word2vec parameter learning explained[J]. arXiv:1411.2738, 2014.
[29] LI J, GALLEY M, BROCKETT C, et al. A diversity-promoting objective function for neural conversation models[C]//Proceedings of the 2016 Conference of the North Ame-rican Chapter of the Association for Computational Lingui-stics: Human Language Technologies, San Diego, Jun 12-17, 2016. Stroudsburg: ACL, 2016: 110-119.