M.R. Boedeltje and A.J. van Hessen
Telecats and Telecats/University of Twente
Abstract
In the past years, the number of service requests through e-mail has shown an explosive growth. Equal to most telephony services handled by call centres, 80% of the incoming e-mails is about 20% of the subjects, making it worthwhile to compose standard answers for at least the 20% most popular questions. Personal answering (each e-mail is answered by a human agent) is simply too expensive to do without the use of predefined answers. By using IR and text classification techniques combined with Natural Language Processing, the process of finding the correct answer for a request can be (partly) automated. In this paper we will describe an e-mail answer suggestion system using IR based classification and NLP techniques. A practical study using an e-mail corpus of 17,000 incoming e-mails (collected and categorized in a Dutch contact centre), has shown that this approach is able to present the correct answer within a ranked list of 5 possible suggestions, for almost 88% of all incoming e-mails. Furthermore, we will show that this approach can be used as well
for spoken content by combining the categorization techniques with the recognition result of the answer on the famous question: ”How can we help you?”.
1 Introduction
With the ongoing acceptance of e-mail as a fast, cheap and reliable way of communication, companies receive an increasing number of service requests via e-mail. To handle these e-mails, call centres are ”transformed” into so-called contact centres handling both e-mail and telephone calls. Since most service requests cover a relatively small set of problems or questions (80% of the mails is about 20% of the subjects), many of these requests may be answered using a relatively small set of standard answers. Handling great amounts of e-mail in a contact centre is a very labour-intensive task, requiring a serious investment of time and money. Automating the answering process could therefore account for serious cost reduction and a decrease in response time. Due to the difficulty of automatically selecting the correct answer and thus the risk of sending the incorrect answer to a customer, most companies are reluctant to incorporate such an automatic e-mail answering system in their contact centres. However, automatically suggesting one or more relevant answers to incoming messages provides a good alternative. If we manage to suggest the correct answer in for instance a top-x of relevant answers, the agent only needs to select the correct answer out of these x answers, instead of
formulating the answer manually or searching the correct answer from all possible answers. Such a system would improve the efficiency in a contact centre and reduce the time spent on answering e-mail. However, one has to tune the number of suggested answers. Suggesting a small number of answers decreases the time an agent has to spend on browsing through the suggestions if the right answer is in these suggestions. However, if the right answer is not present in the suggestions, an agent has to spend extra time on manually searching the correct suggestion. Increasing the number of suggestions increases the chance that right answer is in the
suggestions, but also increases the time an agent has to browse these suggestions. Ideally, no more than 5 suggestions are given to the agents.