Information retrieval based call classification

摘要

In this paper we describe a fully automatic call classification system for customer service selection. Call classification is based on one customer utterance following a "How may I help you" prompt. In particular, we introduce two new elements to our information retrieval based call classifier, which significantly improve the classification accuracy: the use of a-priory term relevance based on class information, and classification confidence estimation. We describe the spontaneous speech recognizer as well as the classifier and investigate correlations between speech recognition and call classification accuracy.