Globenet and RAISE: Intelligent Agents for Networked Newsgroups and Customer Service Support

Abstract

The Globenet system, developed at IBM T.J. Watson Research Center, performs knowledge-based retrieval and handling of newsgroup-like information from heterogeneous WAN sources. Users specify personal, rule-based, intelligent agents that control the retrieval and handling done on their behalves. The work-in-progress version1 of Globenet is enhanced by piloting the RAISE class library, also developed at IBM Watson. RAISE stands for Reusable Agent Intelligence Software Environment. RAISE enables powerful rules and flexible inferencing, as well as smoothly integrated and extensible control of additional actions and tests. Reasoning is performed about both structured and unstructured attributes of newsgroup items/mail-messages. Structured attributes include those available from the message headers, e.g., author and subject. Unstructured attributes include those available from message bodies as well, e.g., free-text conditions of the kind used in classical information retrieval, such as boolean expressions in and phrases. Retrieval/filtering is distributed in space and time: a first, foraging phase at remote sites is followed by a second local phase. Handling functions are also interleaved in this second phase, including categorization, attentional prioritization, visual presentation, selective forwarding and dissemination, and personal information management. Globenet is currently deployed in a customer service support application within IBM to aid human staff in their answering of customer questions and problems. Globenet includes features distinctive to customer service support: it automates identification of questions and supports humans' collaboration in the process of query-answering. Major productivity improvements of over 30% are reported in early field experience with the application. In addition, the approach embodied by the Globenet+RAISE system is applicable to a wide range of other applications in which information has a similar combination of structured and unstructured attributes, e.g., retrieval and handling of e-mail or textual articles, including customized electronic news services.