A Composite Kernel Approach for Dialog Topic Tracking with Structured Domain Knowledge from Wikipedia

Seokhwan Kim, Rafael E. Banchs, Haizhou Li

Abstract: Dialog topic tracking aims at analyzing and maintaining topic transitions in ongoing dialogs. This paper proposes a composite kernel approach for dialog topic tracking to utilize various types of domain knowledge obtained from Wikipedia. Two kernels are defined based on history sequences and context trees constructed based on the extracted features. The experimental results show that our composite kernel approach can significantly improve the performances of topic tracking in mixed-initiative human-human dialogs.