Using Dialogue Acts to Suggest Responses in Support Services via Instant Messaging

Proceedings of the 2006 Australasian Language Technology Workshop (ALTW2006), pages 159–160. Using Dialogue Acts to Suggest Responses in Support Services via Instant Messaging Edward Ivanovic Department of Computer Science and Software Engineering University of Melbourne edwardi@csse.unimelb.edu.au Abstract Instant messaging dialogue is real-time, text-based computer-mediated communi- cation conducted over the Internet. Mes- sages sent over instant messaging can be encoded We propose a method of using dialogue acts to predict utterances in task- oriented dialogue. Dialogue acts provide a semantic representation of utterances in a dialogue. An evaluation using a dia- logue simulation program shows that our proposed method of predicting responses provides useful suggestions for almost all response types. 1 Introduction Support services in many domains have traditionally been provided over the telephone: when customers have queries, they dial a support number and speak to a support representative. Recent years have seen an increasing trend in support services provided over the Internet. Many companies have web sites with Frequently Asked Questions (FAQs), and also of- fer e-mail support. More recently, real-time support via online chat sessions is being offered where cus- tomers and support representatives type short mes- sages to each other. Chat sessions are conducted over a network, such as the Internet, where textual messages can be sent and received between interlocutors in real-time. These chat sessions are commonly referred to as in- stant messaging. Support services that are conducted via instant messaging vary from being person-person dialogue, Speaker Message Agent [Hello Jim]CONVENTIONAL-OPENING, [thank you for contacting MSN Shopping]THANKING. [This is Sanders and I look forward to assisting you today]STATEMENT Agent [How are you doing today?]OPEN-QUESTION Customer [good]STATEMENT, [thanks]THANKING Agent [How may I help you today?]OPEN-QUESTION Table 1: An example of the beginning of a dia- logue in our corpus showing utterance boundaries and dialogue-act tags in superscript. similar to traditional call centres, through to be- ing entirely automated where customers engage in dialogue with a computer program. Commercial software is available to partially automated online support by suggesting responses to a human agent, which may then be accepted or overwritten. The research presented in this paper aims to pro- vide a degree of natural language understanding to assist in automating task-oriented dialogue, such as support services, by suggesting utterances during the dialogue. We apply various probabilistic methods to improve discourse modelling in the support services domain. In previous work, we collected a small corpus of task-oriented dialogues between customers and sup- port representatives from the MSN Shopping online support service (Ivanovic, 2005b). The service is de- signed to assist potential customers with finding var- ious items for sale on the MSN Shopping web site. A sample from one of the dialogues in this corpus is shown in Table 1. The research presented here advances previous work which examined various models and tech- 159 niques to predict dialogue acts in task-oriented in- stant messaging. In Ivanovic (2005b), the MSN Shopping corpus was collected and a gold standard produced by labelling the utterances with dialogue acts. Probabilistic models were then …

Content truncated to 500 words