An Operator Assisted Call Routing System

An Operator Assisted Call Routing System Chun-Jen Leel' 2� Jason S. Chang 2 1 Telecommunication Labs.� Chunghwa Telecom Co., Ltd.�2 Department of Computer Science National Tsing Hua University Chung-Li, Taiwan, R.O.C.� Hsin-chu, Taiwan, R.O.C. cjlee@cht.com.tw � j schang @cs.nthu.edu.tw Abstract A system to assist call routing task for telephone operators at the Directorate General of Telecommunications (DGT) in Taiwan is reported in this paper. The system was developed based on DGT organization profile with description of its six divisions instead of a corpus of recorded and transcribed call-routing dialogs. An acoustic module and an information retrieval module were built specifically for this task. The construction of IR module was based on term extraction and thesaurus discovery processes. By integrating acoustic and IR module, the system achieves satisfactory performance and provides a promising approach to call routing. Simulation results indicated that the proposed algorithm outperforms standard classification methods. A working system based on the proposed approach has been implemented and experimental results are presented. 1� Introduction We proposed an end-to-end system for operator assisted call routing. The system was designed based on the idea that a telephone operator is capable to capture the intent of caller's inquiry but the operators may not be familiar or kept up to date with all the business of the organization. Therefore, he or she can then operate the call router via a speech interface. The system transforms the natural speech, matches the information related to directory, and finally determines the routing destination. The output is then displayed on the screen for the operator to carry out the actual routing action. The goal is to assist the operator in selecting the desired destination that matches caller' s intension. This will expedite operator's response since there is no need to search a printed directory or enter the query in Chinese text. Numerous previous works on call routing has been reported (Riccardi et. al., 1997; Lee et al., 1998; Chu-Carroll and Carpenter, 1999). Most approaches in the literature require a corpus of routed calls to train a routing matrix or language model. Such corpora of dialog are sometimes difficult to obtain due to the concerns over invasion of privacy right. In addition to the acoustic module, a call router also needs to classify caller' s request according to routing destination. This task of classification is similar to text categorization (Lewis et al., 1996) in IR or topic identification (McDonough et al., 1994) in speech research. When a corpus of calls is available, researchers tend to adopt a vector space model, under which terms in the corpus is weighted statistically and treated as independent. Many systems adopted a form-based approach that is commonly used in spoken understanding system (Seneff, Glass, and Goddeau, 1991; Lamel, 1998; Chu-Carroll, 1999; Papineni, Roukos, and Ward, 1999). A form consists of a set of keywords all (or most) of which need to be present in the caller' s request in order for the destination to be activated. This idea amounts to formulating an AND …

Content truncated to 500 words