201607
- Smart Reply: Automated Response Suggestion for Email
- The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems
Smart Reply: Automated Response Suggestion for Email
- Response selection: find the approximate best responses. LSTM model
- Response set generation: deliver high response quality. Select responses from response space which is generated offline using a semi-supervised graph learning approach
- Diversity: choose a small set to show to the user that maximize the total utility. We found that enforcing diverse semantic intents is critical to making the suggestions useful
- Triggering model: a feedforward neural network decides whether or not to suggest responses
很多文章都提到把seq2seq当作纯生成模型效果很差,我自己线下的一些实验也证明了这一点,本文seq2seq的训练方法没有什么变化,只是将它用来做检索而不是生成答案。这样一方面能够保证答案质量(语句通顺,没有语法错误等),另一方面能够利用到seq2seq模型从问题到答案的语义映射能力。可以理解为某种形式的语义搜索。