Skeleton-to-Response: Dialogue Generation Guided by Retrieval Memory
Deng Cai The Chinese University of Hong Kong、Yan Wang Tencent AI Lab、Victoria Bi Tencent AI Lab \ANDZhaopeng Tu Tencent AI Lab、Xiaojiang Liu Tencent AI Lab、Wai Lam The Chinese University of Hong Kong、Shuming Shi Tencent AI Lab Work done while DC was interning at Tencent AI Lab.
Abstract
Traditional generative dialogue models generate responses solely from input queries. Such information is insufficient for generating a specific response since a certain query could be answered in multiple ways. Recently, researchers have attempted to fill the information gap by exploiting information retrieval techniques. For a given query, similar dialogues are retrieved from the entire training data and considered as an additional knowledge source. While the use of retrieval may harvest extensive information, the generative models could be overwhelmed, leading to unsatisfactory performance. In this paper, we propose a new framework which exploits retrieval results via a skeleton-to-response paradigm. At first, a skeleton is extracted from the retrieved dialogues. Then, both the generated skeleton and the original query are used for response generation via a novel response generator. Experimental results show that our approach significantly improves the informativeness of the generated responses.
中文速览
闲聊机器人（chit-chat chatbot）的核心难题在于：仅凭用户输入的一句话，模型很难生成有实质内容的回复，因为同一个问题可以有无数种答案，训练好的模型往往只会输出"我不知道""我觉得是这样"之类安全但无聊的万能句。为此，研究者尝试用检索（information retrieval）技术从训练数据中找出相似对话作为额外参考，但直接把检索结果喂给生成模型会引入大量噪音，模型容易变成简单的"复读机"。本文提出"骨架到回复（skeleton-to-response）"新框架：先从检索到的回复中自动抽取一个骨架——保留有用词、屏蔽与当前问题无关的词——再让生成模型结合原始问题和这个骨架来生成最终回复；为了处理骨架提取过程不可微分的问题，作者分别设计了强化学习（policy gradient）和多任务联合训练两种训练策略。实验结果表明，该方法生成的回复在信息量和针对性上显著优于现有基线模型，为如何高效、可控地将检索结果融入生成式对话系统提供了新思路。
原文 arXiv:1809.05296;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1809.05296v5