作者: Ruhi Sarikaya , Xiaohu Liu
DOI:
关键词:
摘要: A dictionary used by a spoken language understanding (SLU) system is improved providing weightings for entities in the that represent likelihood each entity belongs to an class represented dictionary. classifier model may be trained using seed list containing sample belong and background samples do not class. Clicked URLs from search logs, result URLs, attributes graph as features of train model. The weight candidate are generate use SLU system.