Dialogue Management in Vector-Based Call Routing

Dialogue Management in Vector-Based Call Routing Jennifer Chu-Carroll and Bob Carpenter Lucent Technologies Bell Laboratories 600 Mountain Avenue Murray Hill, NJ 07974, U.S.A. E-mail: {jencc,carp} @ research.bell-labs.corn Abstract This paper describes a domain independent, automat- ically trained call router which directs customer calls based on their response to an open-ended "How may ldi- rect your call?" query. Routing behavior is trained from a corpus of transcribed and handorouted calls and then carried out using vector-based information retrieval tech- niques. Based on the statistical discriminating power of the n-gram terms extracted from the caller's request, the caller is 1) routed to the appropriate destination, 2) trans- ferred to a human operator, or 3) asked a disambigua- tion question. In the last case, the system dynamically generates queries tailored to the caller's request and the destinations with which it is consistent. Our approach is domain independent and the training process is fully automatic. Evaluations over a financial services call cen- ter handling hundreds of activities with dozens of desti- nations demonstrate a substantial improvement on exist- ing systems by correctly routing 93.8% of the calls after punting 10.2% of the calls to a human operator. 1 Introduction The call routing task involves directing a user's call to the appropriate destination within a call center or pro- viding some simple information, such as loan rates. In current systems, the user's goals are typically gleaned via a touch-tone system employing a rigid hierarchical menu. The primary disadvantages of navigating menus for users are the time it takes to listen to all the options and the difficulty of matching their goals to the options; these problems are compounded by the necessity of de- scending a nested hierarchy of choices to zero in on a particular activity. Even simple requests such as "I'd like m), savings account balance" may require users to nav- igate as many as four or five nested menus with four or five options each. We have developed an alternative to touch-tone menus that allows users to interact with a call router in natural spoken English dialogues just as they would with a human operator. Human operators respond to a caller request by 1) routing the call to an appropriate destination, or 2) query- ing the caller for further information to determine where to route the call. Our automatic call router has these two options as well as a third option of sending the call to a human operator. The rest of this paper provides both a description and an evaluation of an automatic call router driven by vector-based information retrieval techniques. After introducing our fundamental routing technique, we focus on the disambiguation query generation module. Our disambiguation module is based on the same sta- tistical training as routing, and dynamically generates queries tailored to the caller's request and the destina- tions with which it is consistent. The main advantages of our system are that 1) it is domain independent, 2) it is trained fully automatically to both route and disam- biguate requests, …

Content truncated to 500 words