Authors: (1) Yugen Sato, Meiji University; (2) Taisei Nakajima, Meiji University; (3) Tatsuki Kawamoto, Meiji University; (4) Tomohiro Takagi, Meiji University. Table of Links Abstract & Introduction Related works Method Experiment Conclusion & References 2 Related works 2.1 Task-Oriented dialogue without a large language model Yichi Zhang et al. [ introduce a novel data augmentation approach and model architecture for generating multiple appropriate responses in task-oriented dialogue systems. The main contribution is to exploit the one-to-many mapping between dialogue states and valid system actions. Focusing on task-oriented dialogue systems that can generate multiple appropriate responses under the same context, we proposed a Multi-Action Data Augmentation (MADA) framework MADA uses training data to generate a one-to-many mapping from dialogue states to valid system actions by MADA is trained to discover all possible mappings from dialogue states to valid system actions, thereby enabling it to generate diverse and appropriate dialogue responses. Zhang_Ou_Yu_2020] 2.2 Task-Oriented dialogue Lang Cao [1] introduces a new method called Dialogue in Diagnosis GPT (DiagGPT) that extends large-scale language models (LLMs) to task-oriented dialogue (TOD) scenarios. The main focus is to improve dialogue in complex diagnostic scenarios such as legal and medical consultations where simple question-answer dialogue is not sufficient. The approach aims to guide the user toward specific task completion, which is a hallmark of TOD, by having the AI chat agent proactively ask questions. Task Guidance: Guides the user to a specific goal by sequencing predefined topics and assists in accomplishing the task through the progression of the dialogue. Proactive Questioning: Gather necessary information from the user by proactively asking questions based on a predefined checklist. Topic Management: Automatically manage topics in dialogue, track topic progression, and effectively participate in discussions around the current topic. Highly scalable: DiagGPT is designed to be flexible enough to incorporate additional features to handle tasks in complex scenarios or to accommodate more needs in the conversation system. DiagGPT is presented as a multi-agent AI system with automatic topic management capabilities to enhance its usefulness in task-oriented dialogue scenarios. Through this design, we aim to better simulate real medical and legal professionals and provide a more intelligent and professional chatbot experience. In the meantime, we will test whether this topic management and task guiding functionality can be provided by a single module, the controller3.1. This paper is under CC 4.0 license. available on arxiv

Task-Oriented Dialogue Systems Beyond Large Language Models

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

102 Languages, One Model: The Multimodal AI Breakthrough You Need to Know

100 Days of AI Day 7: Building Your Own ChatGPT with Langchain

100 Days of AI Day 4: Maximizing Productivity & Creativity with ChatGPT

100 Days of AI Day 6: Retrieval Techniques and Their Use Cases

10 Tips to Get the Most out of ChatGPT

100 Days of AI Day 2: Enhancing Prompt Engineering for ChatGPT

102 Languages, One Model: The Multimodal AI Breakthrough You Need to Know

100 Days of AI Day 7: Building Your Own ChatGPT with Langchain

100 Days of AI Day 4: Maximizing Productivity & Creativity with ChatGPT

100 Days of AI Day 6: Retrieval Techniques and Their Use Cases

10 Tips to Get the Most out of ChatGPT

100 Days of AI Day 2: Enhancing Prompt Engineering for ChatGPT

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps