Knowledge Extraction and Reuse within "Smart" Service Centers

In this paper, we describe the initial version of a text analytics system under development and use at Cisco, where the objective is to "optimize" the productivity and effectiveness of the service center. More broadly, we discuss the practical needs in industry for developing powerful "Smart" Service Centers and the gaps in research to meet these needs. Ideally, service engineers in service centers should be utilized to handle issues which have not been solved previously and machines should be used to solve problems already solved, or at least help the service engineers obtain pertinent information from related and solved service cases when responding to a new request. Such a role for a machine would be a core element of the "Smart Services" offering. Hence, design of a highly efficient human-machine combination to derive insights from text and respond to a user request, is critical and fundamental, this enables service agents to capture relevant information quickly and accurately, and to develop the foundation for upper layer applications. Despite extensive earlier literature, the optimization for service process that involves very long, unstructured documents referencing a number of technology and product related terms with implicit inter-relationships has not been fully investigated. Our approach enables firms such as Cisco to achieve efficient service delivery by automating knowledge extraction to support "Self Service" by end users. The Cisco text analytics system termed Service Request Analyzer and Recommender (SRAR) addresses gaps in the Support Services function, by optimizing the use of human resources and software analytics in the service delivery process. The Analyzer is able to handle complex service requests (SRs) and to present categorized and pertinent information to service agents, based on which the Recommender, an upper layer application, is built to retrieve similar solved SRs, when presented with a new request. Our contributions in the context of text analysis and system design are three-fold. First, we identify the elements of the diagnostic process underlying the creation of SRs, and design a hierarchical classifier to decompose the complex SRs into those elements. Such decomposition provides specific information from the functional perspectives about "What was the problem?" "Why did it occur?" and "How was it solved?" which assists service agents in acquiring the knowledge they need more effectively and rapidly. Second, we build an SR Recommender on top of SR Analyzer to extend the system functionality for improved knowledge reuse, to measure SR similarity for more accurate recommendation of SRs. Third, we validate our SRAR in an initial pilot study in the service center for Cisco network diagnostics and support, and demonstrate the effectiveness and extensibility of our system. Our system appears applicable to the service centers across multiple domains, including networks, aerospace, semiconductors, automotive, health care, and financial services, and potentially adapted and expanded to all the other business functions of an enterprise. We conclude by indicating open research problems and new research directions, to expand the set of problems that need to be addressed in developing a Smart Support Services capability, and the solutions required to achieve them. These include the capture, retrieval, and reuse of more refined, structured and granulated knowledge, as well as the use of forum threads and semi-automated, dynamic categorization, together with considerations of the optimal use of humans and machine learning based software. Other aspects we discuss include recommendation systems based on temporal pattern clustering and incentives for experts to permit their expertise to be captured for machine (re-)use.