【论文阅读】Sequential Matching Network: A New Architecture for Multi-turn Response Selection in Retrieval

2017ACL 论文 作者有来自MSRA的
【论文阅读】Sequential Matching Network: A New Architecture for Multi-turn Response Selection in Retrieval

chatbot 中利用到多轮对话中上下文信息,答案是检索得到的,文章重点在讲如何对多轮对话上下文信息进行建模,答案候选抽取不是重点。

一个示例,比如在下图中两个候选中选哪个?显然应该是候选1,有上下文信息。
【论文阅读】Sequential Matching Network: A New Architecture for Multi-turn Response Selection in Retrieval

模型架构
【论文阅读】Sequential Matching Network: A New Architecture for Multi-turn Response Selection in Retrieval

实现结果
【论文阅读】Sequential Matching Network: A New Architecture for Multi-turn Response Selection in Retrieval

语料:
1. Ubuntu Corpus[1]
2. Douban Conversation Corpus

文章的语料与源码
github:
https://github.com/MarkWuNLP/MultiTurnResponseSelection