Estimating callers' levels of knowledge in call center dialogues

In call centers, since callers range from novices to experts with respect to what they are asking about, callers should be treated differently depending on their knowledge levels. To extract dialogues of callers with a certain knowledge level for an analysis, we propose a method to estimate callers’ levels of knowledge. We focus on features related to vocabulary, utterance timing and duration, and information exchange and use a machine learning technique to learn a classifier that distinguishes the knowledge levels. Experimental results show that our method achieves a precision of 0.8 or better while retaining a moderate recall of around 0.5 in extracting the dialogues of novice callers.