Zero-shot learning of user intent understanding by Convolutional Neural Networks

User intent is a goal that underlies a user-generated utterance which plays a critical role in many intelligent applications, such as dialog systems and search engines. Most previous works focus on intent understanding as a supervised classification problem with the hypothesis that the utterances are labeled in predefined intents. However, how to detect emerging user intents where no labeled utterances are currently tentative. In this paper, we present a zero-shot learning approach for intent understanding, it can predict intents at runtime that did not exist at training time. Our approach extracts semantic features from exiting intents and emerging intents respectively by the Convolutional Neural Networks, and then discriminates emerging intents via knowledge transfer from existing intents. Experiments on a real world dataset show that our model performs better to discriminate emerging intents when no labeled utterances are available.