Towards a framework for collating help-desk responses from multiple documents

Responses to help-desk email inquiries are often repetitive, sharing varying degrees of commonal- ity. In addition, a significant proportion of the re- sponses are generic, containing a very low level of technical content. In this paper, we present a corpus-based approach for identifying common el- ements in help-desk responses and using them to construct a new response. A help-desk domain is unique in that responses that contain even one in- congruous sentence can alienate a user. It is there- fore not always possible to automatically generate a complete response, because personalization is of- ten better handled by human operators. Our system is designed to find and collate the generic portions of responses. We have adapted multi-document summarization techniques, and developed a mea- sure that predicts the completeness of a planned re- sponse, thus indicating when a fully automated re- sponse is possible. Our evaluation shows that 14% of the responses in our corpus can be represented by complete generic responses.