5th International Conference on Spoken Language Processing
We have developed a domain independent, automatically trained, call router which directs customer calls based on their response to an open-ended ``How may I direct your call?'' query. Routing behavior is trained from a corpus of transcribed and hand-routed calls and then carried out using vector-based information retrieval techniques. Terms consist of sequences of morphologically reduced content words. Documents representing routing destinations consist of weighted term frequencies derived from calls to that destination in the training corpus. In this paper, we evaluate our approach in the context of a large financial services call center with thousands of possible customer activities and dozens of routing destinations. We evaluate the system's performance on ambiguous and unambiguous calls when given either accurate transcriptions or fairly noisy real-time speech recognizer output. We conclude that in a highly complex call center, our system performs at roughly the same level of accuracy as human operators.
Bibliographic reference. Carpenter, Bob / Chu-Carroll, Jennifer (1998): "Natural language call routing: a robust, self-organizing approach", In ICSLP-1998, paper 0076.