A Hybrid Ontology Directed Feedback Selection Algorithm for Supporting Creative Problem Solving Dialogues
Hao-Chuan Wang, Rohit Kumar, Carolyn Penstein Rosé, Tsai-Yen Li, Chun-Yen Chang
We evaluate a new hybrid language processing approach designed for interactive applications that maintain an interaction with users over multiple turns. Specifically, we describe a method for using a simple topic hierarchy in combination with a standard information retrieval measure of semantic similarity to reason about the selection of appropriate feedback in response to extended language inputs in the context of an interactive tutorial system designed to support creative problem solving. Our evaluation demonstrates the value of using a machine learning approach that takes feedback from experts into account for optimizing the hierarchy based feedback selection strategy.