Zero-Shot Out-of-Domain Detection

In the field of natural language processing, zero-shot out-of-domain detection is a critical task aimed at identifying unseen texts that come from a different distribution than the training data. The goal of this task is to automatically detect inputs that do not belong to the known domains by learning the characteristics of these domains, thereby enhancing the robustness and generalization capabilities of the system. Its application value lies in effectively filtering out irrelevant or anomalous user requests, improving the accuracy and efficiency of downstream tasks.