AI helps you reading Science

AI generates interpretation videos

AI extracts and analyses the key points of the paper to generate videos automatically


pub
Go Generating

AI Traceability

AI parses the academic lineage of this thesis


Master Reading Tree
Generate MRT

AI Insight

AI extracts a summary of this paper


Weibo:
We propose a new response generation model that learns the post with its set of responses jointly for short-text conversation

Generating Multiple Diverse Responses for Short-Text Conversation.

national conference on artificial intelligence, (2019)

Cited by: 10|Views75
EI
Full Text
Bibtex
Weibo

Abstract

Neural generative models have become popular and achieved promising performance on short-text conversation tasks. They are generally trained to build a 1-to-1 mapping from the input post to its output response. However, a given post is often associated with multiple replies simultaneously in real applications. Previous research on this ta...More

Code:

Data:

0
Introduction
  • Endowing the machine with the ability to converse with humans using natural language is one of the fundamental challenges in artificial intelligence (Turing 1950).
  • Following the most conventional setting for generative short-text conversation models (Shang, Lu, and Li 2015; Li, Monroe, and Jurafsky 2016; Mo et al 2016; Shen et al 2017), the authors consider the single round chi-chat conversation with no context information, i.e. an input post from a user and the output response given by the machine.
  • Despite the popularity of the Seq2seq models, various problems occur when they are applied for short-text conver-
Highlights
  • Endowing the machine with the ability to converse with humans using natural language is one of the fundamental challenges in artificial intelligence (Turing 1950)
  • Following the most conventional setting for generative short-text conversation models (Shang, Lu, and Li 2015; Li, Monroe, and Jurafsky 2016; Mo et al 2016; Shen et al 2017), we consider the single round chi-chat conversation with no context information, i.e. an input post from a user and the output response given by the machine
  • Experimental results show that our model outperforms existing state-of-the art generative methods in generating multiple high-quality and diverse responses
  • We propose a new response generation model that learns the post with its set of responses jointly for short-text conversation
  • Our model can be optimized in a reinforcement learning algorithm, which can deal with the large latent space assumed in our model
  • Our method can effectively increase both the quality and diversity of the multiple generated responses compared with existing baselines and several state-of-the-art generative methods
Methods
  • Beam Search(BS): The authors use the vanilla Seq2seq model with attention employed in decoding (Bahdanau, Cho, and Bengio 2015).
  • Diverse Beam Search(DBS) (Li, Monroe, and Jurafsky 2016): The authors use the same Seq2seq model for training.
  • A modified ranking score is used in beam search to encourage diverse results.
  • Multiple-Mechanism(MultiMech) (Zhou et al 2017) : It introduced latent embeddings for diverse response generation.
  • Following their setting, the authors use 4 latent mechanisms.
  • Ours: The authors explore the model using the pre-trained networks only (Ours(Pretrain)) and two variants utilizing the two loss functions in Eq 5 and 6 in the generation network, denoted as Ours(Avg) and Ours(Min)
Results
  • Results on Weibo

    Overall Performance Results of automatic evaluations and human annotations are shown in Table 1 and 2 respectively.
  • From the human evaluation results, Ours(Min) obtains about 110% improvement over the Seq2seq baselines (BS, DBS and MMI), as well as 70% over the enhanced methods (MultiMech and HGFU).
  • This validates that the method can output more diverse responses while maintaining each of them to be relevant to the post.
  • Full results of all methods are again provided in Appendix
Conclusion
  • The authors propose a new response generation model that learns the post with its set of responses jointly for short-text conversation.
  • The authors' model can be optimized in a reinforcement learning algorithm, which can deal with the large latent space assumed in the model.
  • By sampling multiple diverse latent words from the latent word inference network, the generation network can output different responses.
  • The authors' method can effectively increase both the quality and diversity of the multiple generated responses compared with existing baselines and several state-of-the-art generative methods
Summary
  • Introduction:

    Endowing the machine with the ability to converse with humans using natural language is one of the fundamental challenges in artificial intelligence (Turing 1950).
  • Following the most conventional setting for generative short-text conversation models (Shang, Lu, and Li 2015; Li, Monroe, and Jurafsky 2016; Mo et al 2016; Shen et al 2017), the authors consider the single round chi-chat conversation with no context information, i.e. an input post from a user and the output response given by the machine.
  • Despite the popularity of the Seq2seq models, various problems occur when they are applied for short-text conver-
  • Methods:

    Beam Search(BS): The authors use the vanilla Seq2seq model with attention employed in decoding (Bahdanau, Cho, and Bengio 2015).
  • Diverse Beam Search(DBS) (Li, Monroe, and Jurafsky 2016): The authors use the same Seq2seq model for training.
  • A modified ranking score is used in beam search to encourage diverse results.
  • Multiple-Mechanism(MultiMech) (Zhou et al 2017) : It introduced latent embeddings for diverse response generation.
  • Following their setting, the authors use 4 latent mechanisms.
  • Ours: The authors explore the model using the pre-trained networks only (Ours(Pretrain)) and two variants utilizing the two loss functions in Eq 5 and 6 in the generation network, denoted as Ours(Avg) and Ours(Min)
  • Results:

    Results on Weibo

    Overall Performance Results of automatic evaluations and human annotations are shown in Table 1 and 2 respectively.
  • From the human evaluation results, Ours(Min) obtains about 110% improvement over the Seq2seq baselines (BS, DBS and MMI), as well as 70% over the enhanced methods (MultiMech and HGFU).
  • This validates that the method can output more diverse responses while maintaining each of them to be relevant to the post.
  • Full results of all methods are again provided in Appendix
  • Conclusion:

    The authors propose a new response generation model that learns the post with its set of responses jointly for short-text conversation.
  • The authors' model can be optimized in a reinforcement learning algorithm, which can deal with the large latent space assumed in the model.
  • By sampling multiple diverse latent words from the latent word inference network, the generation network can output different responses.
  • The authors' method can effectively increase both the quality and diversity of the multiple generated responses compared with existing baselines and several state-of-the-art generative methods
Tables
  • Table1: Automatic evaluation results on Weibo. The best/second best results are bold/starred
  • Table2: Human evaluation results on Weibo
  • Table3: Automatic evaluation results on Twitter. The best/second best results are bold/starred
  • Table4: Human evaluation results on Twitter
Download tables as Excel
Related work
  • The Seq2seq framework has been widely used for conversational response generation (Vinyals and Le 2015; Sordoni et al 2015; Shang, Lu, and Li 2015). Such models learn the mapping from an input x to one output y by maximizing the pairwise probability of p(y|x). During testing, these models only target for one response. In order to obtain multiple responses, beam search can be used. However, the resulting multiple sequences are often very similar. Many approaches have been proposed to re-rank diverse meaningful answers into higher positions. For example, Li et al (2016) proposed a simple fast decoding algorithm to directly encourage response diversity in the scoring function used in beam search. Shao et al (2017) heuristically re-ranked the responses segment by segment to inject diversity earlier in the decoding process. These methods only modified the decoding steps and still often generated responses using different words but with similar semantics.
Funding
  • This work was supported by National Natural Science Foundation of China (Grant No 61876120)
Reference
  • Adi, Y.; Kermany, E.; Belinkov, Y.; Lavi, O.; and Goldberg, Y. 2016. Fine-grained analysis of sentence embeddings using auxiliary prediction tasks. arXiv preprint arXiv:1608.04207.
    Findings
  • Bahdanau, D.; Cho, K.; and Bengio, Y. 2015. Neural machine translation by jointly learning to align and translate. Proceedings of the 2015 International Conference on Learning Representations(ICLR).
    Google ScholarLocate open access versionFindings
  • Li, J.; Galley, M.; Brockett, C.; Gao, J.; and Dolan, B. 2016a. 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(NAACL-HLT), 110–119.
    Google ScholarLocate open access versionFindings
  • Li, J.; Monroe, W.; Ritter, A.; Jurafsky, D.; Galley, M.; and Gao, J. 2016b. Deep reinforcement learning for dialogue generation. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing(EMNLP), 1192–1202.
    Google ScholarLocate open access versionFindings
  • Li, J.; Monroe, W.; Shi, T.; Jean, S.; Ritter, A.; and Jurafsky, D. 2017. Adversarial learning for neural dialogue generation. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, 2157–2169.
    Google ScholarLocate open access versionFindings
  • Li, Z.; Jiang, X.; Shang, L.; and Li, H. 2018. Paraphrase generation with deep reinforcement learning. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing.
    Google ScholarLocate open access versionFindings
  • Li, J.; Monroe, W.; and Jurafsky, D. 2016. A simple, fast diverse decoding algorithm for neural generation. arXiv preprint arXiv:1611.08562.
    Findings
  • Liu, C.-W.; Lowe, R.; Serban, I.; Noseworthy, M.; Charlin, L.; and Pineau, J. 2016. How not to evaluate your dialogue system: An empirical study of unsupervised evaluation metrics for dialogue response generation. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, 2122–2132.
    Google ScholarLocate open access versionFindings
  • Mikolov, T.; Chen, K.; Corrado, G.; and Dean, J. 2013. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781.
    Findings
  • Mo, K.; Li, S.; Zhang, Y.; Li, J.; and Yang, Q. 2016. Personalizing a dialogue system with transfer reinforcement learning. arXiv preprint arXiv:1610.02891.
    Findings
  • Papineni, K.; Roukos, S.; Ward, T.; and Zhu, W.-J. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics(ACL), 311–318. Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Shang, L.; Lu, Z.; and Li, H. 2015. Neural responding machine for short-text conversation. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics(ACL) 1577–1586.
    Google ScholarLocate open access versionFindings
  • Shao, L.; Gouws, S.; Britz, D.; Goldie, A.; Strope, B.; and Kurzweil, R. 2017. Generating long and diverse responses with neural conversation models. arXiv preprint arXiv:1701.03185.
    Findings
  • Shen, X.; Su, H.; Li, Y.; Li, W.; Niu, S.; Zhao, Y.; Aizawa, A.; and Long, G. 2017. A conditional variational framework for dialog generation. Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics(ACL).
    Google ScholarLocate open access versionFindings
  • Sordoni, A.; Galley, M.; Auli, M.; Brockett, C.; Ji, Y.; Mitchell, M.; Nie, J.-Y.; Gao, J.; and Dolan, B. 20A neural network approach to context-sensitive generation of conversational responses. In Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies(NAACL-HLT), 196–205.
    Google ScholarLocate open access versionFindings
  • Turing, A. 1950. Computing machinery and intelligence. Mind.
    Google ScholarLocate open access versionFindings
  • Vinyals, O., and Le, Q. 2015. A neural conversational model. arXiv preprint arXiv:1506.05869.
    Findings
  • Wang, L.; Schwing, A.; and Lazebnik, S. 2017. Diverse and accurate image description using a variational auto-encoder with an additive gaussian encoding space. In Advances in Neural Information Processing Systems, 5756–5766.
    Google ScholarLocate open access versionFindings
  • Wen, T.-H.; Miao, Y.; Blunsom, P.; and Young, S. 2017. Latent intention dialogue models. In International Conference on Machine Learning, 3732–3741.
    Google ScholarLocate open access versionFindings
  • Williams, R. J. 1992. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Machine Learning 8(3-4):229–256.
    Google ScholarLocate open access versionFindings
  • Yao, L.; Zhang, Y.; Feng, Y.; Zhao, D.; and Yan, R. 2017. Towards implicit content-introducing for generative short-text conversation systems. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, 2190–2199.
    Google ScholarLocate open access versionFindings
  • Zhao, T.; Lee, K.; and Eskenazi, M. 2018. Unsupervised discrete sentence representation learning for interpretable neural dialog generation. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (ACL).
    Google ScholarLocate open access versionFindings
  • Zhao, T.; Zhao, R.; and Eskenazi, M. 2017. Learning discourselevel diversity for neural dialog models using conditional variational autoencoders. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (ACL), 654–664.
    Google ScholarLocate open access versionFindings
  • Zhou, G.; Luo, P.; Cao, R.; Lin, F.; Chen, B.; and He, Q. 2017. Mechanism-aware neural machine for dialogue response generation. In Proceedings of AAAI Conference on Artificial Intelligence (AAAI), 3400–3407.
    Google ScholarLocate open access versionFindings
  • Zhou, G.; Luo, P.; Xiao, Y.; Lin, F.; Chen, B.; and He, Q. 2018. Elastic responding machine for dialog generation with dynamically mechanism selecting. In Proceedings of AAAI Conference on Artificial Intelligence (AAAI).
    Google ScholarLocate open access versionFindings
  • Zhou, Z.-H. 2004. Multi-instance learning: A survey. Department of Computer Science & Technology, Nanjing University, Tech. Rep.
    Google ScholarLocate open access versionFindings
Your rating :
0

 

Tags
Comments
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn
小科