Dialogue Generation Using Self-Attention Generative Adversarial Network
Document Type
Conference Proceeding
Publication Date
12-9-2019
School
Computing Sciences and Computer Engineering
Abstract
Developing an intelligent conversation system is one of the longest-running goals in Artificial Intelligence research. For a successful conversation system, appropriate dialogue generation is one of the main components. This article describes a method to generate dialogue using Self-Attention Generative Adversarial Network (SAGAN). The general objective of this method is to produce dialogue which is very similar to human-generated dialogue both structurally and contextually. Generative Adversarial Network (GAN) is mainly used to generate dialogue which is structurally very similar to human-generated dialogue. On the other hand, the self-attention network helps to maintain the context of the conversation in extreme detail for both single track and multi-track dialogue generation. SAGAN enhances the accuracy for both single track and multi-track dialogue generation in a significant way.
Publication Title
2019 IEEE International Conference on Conversational Data & Knowledge Engineering
Recommended Citation
Hatua, A.,
Nguyen, T. T.,
Sung, A. H.
(2019). Dialogue Generation Using Self-Attention Generative Adversarial Network. 2019 IEEE International Conference on Conversational Data & Knowledge Engineering.
Available at: https://aquila.usm.edu/fac_pubs/17036