<style>
img {
display: block;
margin-left: auto;
margin-right: auto;
}
</style>

> [Paper link](https://aclanthology.org/2022.acl-long.425.pdf) | [Note link](https://zhuanlan.zhihu.com/p/523732757) | [Code link](https://github.com/MiuLab/SalesBot) | ACL 2022

:::success
**Thoughts**

:::

## Abstract

This paper proposes a framework to automatically generate many dialogues without human involvement, in which any powerful open-domain dialogue generation model can be easily leveraged.

## Introduction

**Open-domain**, which chat with users via suitable, engaging, safe conversations.

**Task-oriented dialogues**, which include plenty of multi-domain dialogues with state information to track users’ behaviors.

Recent work merged those two capabilities by inserting chit-chat sentences into the existing task-oriented dialogue data. The idea is to **allow the agent gains more social, personalized communication skills when focusing on task-oriented dialogue generation.**


## Proposed Approach

### Open-Domain Dialogue Generation

To generate high-quality open-domain dialogues, the pre-trained dialogue generation models can be adopted. Here they choose **BlenderBot.**

They manipulate the user and the sales to have different personas in order to cover wide-range topics in their generated dialogues.

In this part, they use **ParlAI** to build two BlenderBots to self-chat with each other in order to construct various dialogues involving different personas.

### Chit-Chat to Task-Oriented Transition

This paper proposes two components to address two main challenges:
- how to capture the suitable timing
- how to promote the target products/tasks

### Task-Oriented Dialogue Generation

## Data Quality Evaluation

## Results and Analysis

## Related Work

## Conclusion