Autority aware expert search: Algorithm and system for NSFC

This paper describes the peer reviewer finding algorithm used in an aided system to assist the proposal reviewing task in the National Science Foundation of China (NSFC)[1]. We propose a new probabilistic language model in which expert authority is taken into consideration, by introducing a prior probability of candidate into the model. Application codes representing research areas are defined by NSFC and are available for both experts and proposals. We integrate code information into our model as we add a code probability term into the model, thus promote rankings of experts with matched code. All proposals submitted to NSFC are written in Chinese, and since Chinese word segmentation is non-trivial task whose accuracy affects the final results heavily, we try to improve segment accuracy by adding domain-specific terms to a user-editable dictionary. Terminologies are extracted from certain field of bibliography data in NSFC system. Experiments show that our algorithm is effective in real world system like NSFC.