همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

PerSHOP: A Multi-Domain Persian Dataset for Shopping Dialogue Systems

نویسندگان :

Keyvan Mahmoudi¹ Heshaam Faili²

1- دانشگاه تهران 2- دانشگاه تهران

کلمات کلیدی :

task-oriented dialogue،dialogue datasets،shopping dialogues،intent classification،slot filing،persian language

چکیده :

Task-oriented dialogue systems require high-quality datasets that reflect realistic user–agent interactions. Persian, however, lacks resources in this area, especially for commercial and retail applications. We introduce PerSHOP, the first large-scale, multi-domain Persian dataset for shopping-oriented task-based dialogues. PerSHOP contains 1,061 multi-turn buyer–seller conversations grounded in a structured knowledge base of 750 products, 84 sub-domains, and 33 semantic slots. Dialogues were collected through a custom crowdsourcing platform and annotated using a hybrid rule-based and human verification pipeline. We provide benchmark results for intent classification and slot filling using zero-shot large language models and supervised transformer-based models. PerSHOP fills a critical gap in Persian dialogue resources and supports future research in domain-specific conversational AI. The dataset is publicly released for research use.