Combining distance and sequential dependencies in expert finding

Expert finding is the task of identifying persons with expertise on a given topic. Existing methods try to model the dependencies between candidates and terms with distance measure or sequential measure, which have been proven to be effective. However, to the best of our knowledge, no work has been conducted on the combination of the two dependencies. In this paper, we propose a language model based method to combine both dependencies under a unified framework. Specifically, we first propose an order kernel based document representation for incorporating the sequential dependency, and then we combine it with the proximity kernel based document representation which is designed to model the distance dependency. Our experiment results demonstrate the effectiveness of the order kernel and show that a linear combination of both dependencies can improve the performance significantly over the baseline method.