• Publications
  • Influence
Posterior-GAN: Towards Informative and Coherent Response Generation with Posterior Generative Adversarial Network
TLDR
A novel encoder-decoder based generative adversarial learning framework, Posterior Generative Adversarial Network (Posterior-GAN), which consists of a forward and a backward generative discriminator to cooperatively encourage the generated response to be informative and coherent by two complementary assessment perspectives is proposed.
Regularizing Dialogue Generation by Imitating Implicit Scenarios
TLDR
This work proposes to improve generative dialogue systems from the scenario perspective, where both dialogue history and future conversation are taken into account to implicitly reconstruct the scenario knowledge.
Generating Relevant and Coherent Dialogue Responses using Self-Separated Conditional Variational AutoEncoders
TLDR
Self-separated Conditional Variational AutoEncoder (abbreviated as SepaCVAE) is proposed that introduces group information to regularize the latent variables, which enhances CVAE by improving the responses’ relevance and coherence while maintaining their diversity and informativeness.
Multi-View Feature Representation for Dialogue Generation with Bidirectional Distillation
TLDR
A novel training framework that extends the unidirectionaldistillation to the bidirectional distillation that encourages the student and its student peers to co-evolve by exchanging complementary knowledge with each other and improves the model generalization without sacrificing training efficiency.
Collaborative Group Learning
TLDR
Empirical evaluations indicate that the proposed Collaborative Group Learning method significantly outperforms various state-of-the-art collaborative approaches whilst enhancing computational efficiency.
Neural Dialogue Generation Methods in Open Domain: A Survey
  • Bin Sun, Kan Li
  • Computer Science
    Natural Language Processing Research
  • 1 March 2021
Neural Dialogue Generation Methods in Open Domain: A Survey
TLDR
This survey elaborated the research history of these existing generative methods, and roughly divided them into six categories, i.e., Encoder-Decoder framework-based methods, Hierarchical Recurrent EncodedDecoder (HRED) based methods, Variational Autoencoder (VAE)-basedmethods, Reinforcement Learning (RL)based methods, GenerativeAdversarial Network (GAN)-based Methods, and pretraining-model-based Methods.
Continue or SHIFT: Learning Conversational Patterns for Dialogue Generation
TLDR
The loose coupling approach (LCA) is proposed to directly address the learning of such conversational patterns, which includes two parts: a high-level information selecting mechanism that enables the model to ignore the context or the previous part of the response, and a sampling-based response guide mechanism that mitigates the exposure bias problem in training.
Diversifying Neural Dialogue Generation via Negative Distillation
TLDR
A novel negative training paradigm is proposed, called negative distillation, to keep the model away from the undesirable generic responses while avoiding the above problems, and empirical results show that this method outperforms previous negative training methods.
...
1
2
...