Automatically finding experts in large organizations

Abstract

Automatically finding experts is a critical need for distributed organizations managing employees' knowledge. This paper presents an approach that models a probabilistic cascading framework to find relevant experts in enterprise corpora. We employ a qualification of experience that is validated as a measure of expertise. A language model for each experience measure is estimated with topical words. Then for each candidate's expertise, a language model is estimated with its associated measures. Cascading of these models, we can capture how the expertise is relevant to a topical query. Our evaluation on TREC Enterprise corpora shows that this is an effective approach for expert finding. Moreover, its performance could be further improved by clustering of relevant experience measures.