End-to-end model for multi-turn conversational dialogue agent

多轮会话对话代理的端到端模型

• 批准号: 2102871
• 金额: --
• 依托单位: University of Nottingham
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2018
• 资助国家: 英国
• 起止时间: 2018 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2102871
• 关键词: End; end; model; multi; turn

Conversational dialogue has been researched for many years, especially with the advert of deep learning (DL). Researchers tend to introduce DL into dialogue construction. Historically, the approaches for conversational dialogue evolved from rule-based to machine learning based. Due to the popularity of sequence to sequence model, the quality of conversational dialogue improved significantly. However, there are still many challenges that need to be conquered, including multi-turn dialogue, consistent persona, and evaluation method. In this project, I mainly focus on these challenges, aiming at proposing a novel end-to-end model to implement a multi-turn conversational dialogue agent.

会话式对话已经被研究了很多年，特别是随着深度学习（DL）的推广。研究人员倾向于将深度学习引入对话构建中。从历史上看，会话对话的方法从基于规则演变为基于机器学习。由于序列到序列模型的流行，会话对话的质量显着提高。然而，仍有许多挑战需要克服，包括多轮对话、一致的角色和评估方法。在这个项目中，我主要关注这些挑战，旨在提出一种新颖的端到端模型来实现多轮会话对话代理。