PETIS: Intent Classification and Slot Filling for Pet Care Services

During the COVID-19 pandemic, the surge in online pet care services led to an increased demand for conversational AI systems specifically designed for the veterinary domain. However, traditional natural language understanding (NLU) tasks and datasets often fall short due to domain-specific terminology, the descriptive nature of user utterances, and the high cost of expert annotations. To fill this gap, we introduce PETIS, a novel dataset comprising 10,636 annotated utterances specifically designed for intent classification and slot filling in pet care domain, featuring 10 unique intent classes and 11 slot classes. PETIS addresses the scarcity of annotated data in this domain and serves as a challenging benchmark for evaluating NLU models. We demonstrate its effectiveness through experiments using state-of-the-art models, achieving 93.32 accuracy in intent classification and a Micro F1 (c) score of 91.21 in slot filling using multitask AdapterFusion. Furthermore, domain adaptation significantly enhanced performance, showcasing the potential of PETIS to drive research and development in conversational AI for online pet care services, offering a valuable resource for advancing the field.