One way to find information that may be required, is to approach a person who is believed to possess it or to identify a person who knows where to look for it. Technical support, which automatically compiles individual expertise and makes this accessible, may be centred on an expert finder system. A central component of such a system is a user profile, which describes user expertise level in discussed subjects. Previous works have made attempts to weight user expertise by using content-based methods, which associate the expertise level with the analysis of keyword usage, irrespective of any semantic meanings conveyed. This paper explores the idea of using a natural language processing technique to understand given information from both a structural and semantic perspective in building user profiles. With its improved interpretation capability compared to prior works, it aims to enhance the performance accuracy in ranking the order of names of experts, returned by a system against a help-seeking query. To demonstrate its efficiency, e-mail communication is chosen as an application domain, since its closeness to a spoken dialog, makes it possible to focus on the linguistic attributes of user information in the process of expertise modelling. Experimental results from a case study show a 23% higher performance on average over 77% of the queries tested with the approach presented here.
展开▼
